D3 renderer & GUI: file:// CORS fix, branch jitter, default GeoJSON, Summary-tree labels, flat output

[plemey]

This PR bundles several independent renderer/UI improvements (plus the incidental
build fixes needed to compile/verify on a modern JDK). They are independent and
could be split if a reviewer prefers.

---

1. Fix file:// CORS error when loading a visualization

Opening a visualization from disk (file://) failed: the browser blocked the
data.json XHR as a CORS violation. The data is now inlined into main.js
(var SPREAD3_DATA = ...) via a loadData() helper that falls back to fetching
data.json when not inlined; D3Renderer substitutes the input JSON into the
generated main.js instead of copying a separate data.json.

2. Jitter the curvature of overlapping discrete branches

Branches connecting the same two locations were drawn as identical arcs and
superimposed. generateLinesLayer groups branches by their unordered location
pair and gives each a distinct arc side (sweep flag) and radius multiplier
so they fan out. Singletons/continuous branches unchanged. New lineCurvatureJitter.

3. Load a bundled world.geojson by default

data/geoJSON_maps/world.geojson is bundled into the jar and used as the default
for "Load GeoJSON file:". Utils.getDefaultGeojsonPath() resolves it (jar →
extract to temp; IDE → data/; cached). Panels default to it only when null, so a
manual Load overrides it. Note: the field now defaults non-null (a world layer
is always included unless another file is loaded; no in-GUI "none" option yet).

4. Indicate the active GeoJSON on the Load button

So users know a default is already loaded, the "Load GeoJSON file:" button now
shows the active file's name, tagged "(default)" when it is the bundled world map;
the tag drops once another file is loaded. Full path stays in the tooltip.

5. Rename "MCC tree" → "Summary tree" in the GUI

Relabel the user-facing input-/analysis-type strings (summary trees need not be
MCC). Internal enum codes (MCC_TREE, ...) unchanged.

6. Write output directly into the selected folder

Jar mode preserved the renderers/d3/d3renderer/ prefix from jar entry names,
nesting output deeply. That prefix is stripped so index.html/main.js land
directly in the user-selected folder, matching the IDE path.

7. Fix branches painted short of the location after increasing curvature

The curvature slider recomputed each branch's stroke-dasharray (reveal mask)
with getTotalLength() measured at the start of the transition — against the
old, shorter arc. The mask was then too short for the new, longer arc, leaving the
segment near the target unpainted (branch appears to stop short). The mask is now
recomputed on the transition end event from the final morphed length.

9. Bump gson 2.3.1 -> 2.11.0 and commons-io 2.4 -> 2.16.1

Both were ~2014 releases; upgraded to current stable versions (drop-in, no API
changes for the calls used). Updated the bundled jars, the build.xml includes and
the Eclipse .classpath. Verified end-to-end: commons-io drives the renderer
file inlining and gson parses the bundled world.geojson against the new jars.

10. Migrate Joda-Time to java.time (drop the dependency)

Joda-Time (2014) is superseded by the JDK's built-in java.time. Migrated all
usages across TimeParser, DateEditor, JsonMerger and KmlRenderer
(formatter/parse/print, LocalDate.of, getMonthValue, Date→LocalDate, and
KmlRenderer's interval/Duration math) and removed joda-time-2.7.jar. Patterns
use uuuu (proleptic year) not yyyy (year-of-era), so the existing
negative/zero-year (ancient date) support is preserved. Verified TimeParser
round-trips slash, decimal and negative-year inputs (e.g. -100.5 → -0100/06/01).

11. Modernize the Swing look & feel (FlatLaf + light/dark toggle)

The desktop app rendered in a dated Metal/Nimbus look (on macOS it tried the
abandoned, unbundled Quaqua L&F and fell back to the default). FlatLaf
(FlatLightLaf) is now the primary look & feel, set before any Swing components
are created; the previous Quaqua/Nimbus/system-default attempts are kept as
fallbacks (gated behind !lafLoaded), and the FlatLaf call is guarded with
Throwable so a missing jar degrades gracefully. A new View menu adds a
Dark theme checkbox that swaps FlatLightLaf <-> FlatDarkLaf and calls
FlatLaf.updateUI() to restyle all open windows live. Bundles flatlaf-3.7.1.jar.
Independent of the deferred d3 renderer work.

12. Default the discrete location attribute to "location" / "loc"

The "Select location attribute" combo (discrete tree panel) previously defaulted
to the first annotation name. It now selects location by default, or loc as a
second choice, when present in the tree's annotation list (otherwise the first
item is kept). The selection is applied on the EDT in the tree-load worker's done() (after the
selector is built and added), so it behaves exactly like a manual pick: the
"Setup location coordinates" component appears underneath and
settings.locationAttributeName is set, with no second selection required.

13. Fix the Setup-locations window not rendering under FlatLaf

The location coordinates editor built its top button bar from a JMenuBar used
as a generic container (added to the dialog's NORTH, never set as the window's
menu bar). Under FlatLaf the menu-bar UI assumes real menus, so the panel inside
it was not painted until a hover/click forced a repaint (the previous L&F
tolerated the misuse). Replaced it with a plain JPanel -- same layout, no
behaviour change. This was the only such misuse in the codebase.

14. Narrow the default main window (1400 -> 800)

The main window opened at 1400x700, leaving a lot of empty horizontal space. The
default width is now 800 (height unchanged); the window stays resizable and the
300x200 minimum is unchanged.

Incidental: unblock the build on modern JDKs

• build.xml: javac source/target 1.6 → 1.8.
• Removed two dead, non-portable internal-JDK imports (com.sun.corba...Util in
  ContinuousTreeParser, com.sun.org.apache.xerces.internal...SynchronizedSymbolTable
  in MergePanel) that were unused and break/warn on newer JDKs.

Testing

• ant build → BUILD SUCCESSFUL; jar contains patched main.js and
  geojson/world.geojson.
• node --check passes on the template and a simulated inlined bundle.
• Branch-jitter side/radius assignment verified via isolated logic test.
• Default GeoJSON verified end-to-end (extract-from-jar → parseGeoJSON()).
• Flattened output verified end-to-end (headless render() writes
  <output>/index.html + main.js with data inlined).

8. Resolve the running jar by absolute path

getRunningJarName() returned only the jar's basename, so new JarFile(name)
resolved it against the working directory — rendering only worked when the app
was launched from the jar's own directory and threw NoSuchFileException
otherwise. The absolute path is now parsed from the jar:file:/...!/... URL
(via URI, so encoded chars are handled), with a string fallback. Verified: a
headless render() now succeeds from an unrelated working directory.

Follow-up (deferred): d3 v3.5 -> v7 renderer upgrade

The renderer bundle (src/renderers/d3/d3renderer/main.js) embeds d3 v3.5.17
and jQuery v1.4.2. Upgrading is deliberately out of scope for this PR because
it is a standalone project, not an in-repo edit:

• Source lives elsewhere. The bundle is built from
  phylogeography/d3-renderer
  (a webpack project, last touched Feb 2017). This repo contains only the built
  artifact. The app code to migrate is ~120 KB across ~11 modules
  (lines / topo / points / areas / counts / main / locations / time / ...), plus
  three vendored components (d3-legend, jquery.simple-color, d3.slider).
• Two coupled major migrations. d3 v3 -> v7 (four majors; the app uses the
  full v3 API: d3.scale.*, d3.geo.*, d3.svg.*, d3.layout.hierarchy,
  d3.behavior.drag, callback-style d3.json) and webpack 1.9 -> 5 (config
  • loader ecosystem rewrite; some deps install via git+ssh).
• Dead vendored pieces. d3.slider has no v4+ release (replace or rewrite);
  d3-svg-legend changed APIs.
• No longer user-facing. The inline-data change (order in continuous tree distribution analysis #1) already removed the
  file:// CORS failure, so the remaining motivation is maintainability.

Recommended approach: a separate PR against phylogeography/d3-renderer (port the
modules onto a fresh webpack 5 + d3 v7 skeleton, replace the dead vendored
components), with visual regression testing across all four output types
(discrete, continuous, time-slicer, bayes), then drop the rebuilt bundle into
SpreaD3.