feat: migrate publishDir to workflow output definition (pilot, 15/19)

[qclayssen]

Summary

Pilots migrating sash's publishDir blocks to Nextflow's new workflow output definition (publish: section + top-level output {} block), which came out of preview and is stable as of Nextflow 25.10.0 — no feature flag needed on the 26.04.6 this branch already runs (built on top of #64). publishDir is being deprecated in favor of this.

Migrates 15 of sash's 19 publishDir blocks in conf/modules.config. Verified path-identical to the current publishDir/saveAs output layout via ~10 standalone empirical test pipelines, since the new syntax's directory/subdir-prefix semantics aren't well documented and differ subtly from legacy saveAs.

Base branch is refactor/nextflow-strict-syntax (#64) since this depends on the Nextflow 26.04.6 bump — rebase onto main once that merges.

What changed

• conf/modules.config: removed publishDir from the 15 migrated withName: blocks.
• main.nf: added the publish: section and a new top-level output {} block on the entry workflow {} (Nextflow only permits publish: in the unnamed entry workflow).
• workflows/sash.nf / main.nf (UMCCR_SASH): added emit: blocks passing the 18 named channels needed by the publish paths up from the subworkflows to the entry workflow.
• subworkflows/local/{purple_calling,linx_annotation,linx_plotting}.nf: added small additive _raw emits alongside each subworkflow's existing emit, exposing the process-scoped meta (needed for meta.key in publish paths). No logic changes.
• nextflow.config: workflow.output.mode = params.publish_dir_mode, the new system's equivalent of the old per-block mode:.

What's deliberately NOT migrated (4 of 19)

BOLT_SMLV_SOMATIC_REPORT, BOLT_SMLV_GERMLINE_REPORT, BOLT_OTHER_CANCER_REPORT, BOLT_OTHER_MULTIQC_REPORT remain on publishDir in conf/modules.config, unchanged. All four do filename-pattern-based routing/renaming inside their saveAs closure — e.g. BOLT_SMLV_SOMATIC_REPORT pulls one specific *.pcgr.grch38.html out of a directory of PCGR output files and renames it to ${meta.key}/${meta.tumor_id}.pcgr.html, while everything else in that same process's output goes to a different subpath (${meta.key}/smlv_somatic/report/). BOLT_SMLV_GERMLINE_REPORT does the same for CPSR HTML; BOLT_OTHER_CANCER_REPORT/BOLT_OTHER_MULTIQC_REPORT do it for the cancer/multiqc reports.

The new workflow-output system's path {} closures operate on a channel's structured value (meta, file ->), not on arbitrary filenames matched against a directory the way saveAs does. Reproducing this exactly requires splitting each of these 4 processes' own output: block into separate named emit channels first (e.g. one channel for "the html report", one for "the rest of the output directory") — a change to the module .nf files themselves, not just config/workflow wiring. That's a materially bigger and riskier change than this pilot (touches process-level output declarations for report-generating processes whose output is what clinicians/curators actually review), so it's left as follow-up work, tracked here rather than attempted in this PR.

publishDir remains fully functional (deprecated, not removed) in the meantime, so nothing breaks by leaving these 4 as-is.

Verification

• nextflow lint . -exclude .git -exclude work -exclude .nf-test -exclude nf-test.config -o concise — 0 errors, 0 warnings.
• nextflow run . --help — parses cleanly.
• NXF_SYNTAX_PARSER=v1 nf-test test — 11/13 passing, matching the known-good baseline on refactor: Nextflow 26.04.6 + strict syntax migration #64 exactly (2 pre-existing failures, unrelated).
• nextflow run . -profile test,docker -stub-run — compiles the new publish:/output {} wiring with no structural error (fails only on the pre-existing conf/test.config missing-fixture gap tracked in refactor: Nextflow 26.04.6 + strict syntax migration #64, not on anything from this PR).
• Not yet done: no real-data run to visually confirm the published directory tree matches today's — the empirical test pipelines proved the syntax patterns in isolation, but a real end-to-end run against a test sample (see docs/testing-validation.md tier 3) is the actual sign-off before this merges.

Follow-up (separate work, not this PR)

• Split the 4 report-generating processes' output declarations to migrate the remaining publishDir blocks.
• Real-data validation run comparing this branch's output tree to main's, byte-for-byte on the migrated paths.

[qclayssen]

Pushed `652c433` applying the confirmed simplification from the review: dropped the 4 redundant `_raw` emit channels (`purple_dir_raw`, `somatic_raw`, `germline_raw`, `plot_dir_raw`).

Cross-checked against Nextflow's own docs/tutorial for the workflow output system before making this change — there's no precedent there for a parallel "raw meta" channel; closures just read whatever field the channel's existing meta carries. Combined with the earlier empirical proof that `restoreMeta()`'s join means the restored channel's `meta.id` always equals what `meta.key` would be on the raw channel, this confirmed the simplification was safe: switched `main.nf`'s `purple_dir`/`linx_germline_annotation_dir`/`linx_somatic_annotation_dir`/`linx_somatic_plot_dir` path closures to read `meta.id` off the already-existing restored channels instead.

Also renamed `linxreport_raw` → `linxreport` for clarity — that one wasn't actually redundant (LINXREPORT's output never goes through `restoreMeta()`, so it was always the only channel available), just misleadingly named after the same pattern.

`nextflow lint`: 0/0. `nf-test`: 11/13 (unchanged). `--help` parses clean.