crash fix: sync stale editor cursors through document updates

[Third-Thing]

Problem: When multiple editors share a TextDocument, or when a document is edited programmatically, only the initiating editor updates its cursor. Other editors keep stale byte offsets. That can leave a cursor past the end of the buffer or inside a multibyte character, which then panics on the next keypress. In practice this affects both shared editors and single editors after raw Document::edit / edit_single calls.

Solution: Register a TextDocument update listener for each Editor and remap non-originating cursors through incoming RopeDeltas with Cursor::apply_delta. The originating editor is skipped to avoid double-transforming its cursor. The listener is registered when the editor is created and when its document changes, and it uses try_update plus a document identity check so disposed editors or stale listeners fail closed instead of panicking.

Reasoning: This is the smallest change that restores cursor validity using floem’s existing document update flow and eliminates the crashes mentioned in #934 & #693. TextDocument already emits deltas for every edit, and Cursor already knows how to transform itself through those deltas. Hooking editors into that path keeps cursor synchronization reactive and document-driven, without introducing new edit APIs or a larger coordination layer. The document identity guard is necessary because add_on_update is currently append-only, so old listeners cannot yet be removed when an editor switches documents.

Associated with issue #1058