[raiseCatError]

Stacked on #411 (design doc). Merge #411 first; GitHub then retargets this PR to dev.

What and why

#403: Fish test teardown race (test fixture, not a product leak). On a home with no generated completions, an interactive Fish starts fish_update_completions detached:

• Fish 4.x checks $__fish_cache_dir/generated_completions, Fish 3.x $__fish_user_data_dir/generated_completions.
• The orphaned Python process writes about 1,400 files into the test home after the shell exits, which races rmSync (the CI ENOTEMPTY).
• Production keeps Fish's normal one-time behavior with the person's own directories.

Fix:

• scripts/test-shell-home.mjs marks that generation as done in test homes. The runner, LiveSandbox and the fixtures that start Fish with their own home use it.
• The runner no longer passes ambient XDG_DATA_HOME/XDG_CACHE_HOME/XDG_STATE_HOME to tests. Under a wrapper that sets a shared cache, test shells had been writing there.
• No retries were added.

#408: queued-highlight classification (a production race, plus a test race).

• Every prompt marker's shell knowledge cleared SemanticService's cache and resolved in-flight lookups as unknown. A stored command whose lookup was in flight when the next prompt arrived (a fast queue) kept the unclassified color for good.
• A superseded lookup is now asked again against the new knowledge, at most 3 times. A stale answer is still never cached.
• The test had also raced the real shell's startup knowledge. It now waits for that knowledge and for the lookups it started, instead of one setImmediate.

#388: multi-row progress redraws.

• AnsiOutputParser handled \r and current-line erases but ignored row movement, so each frame of a multi-row progress display became permanent lines.
• Programs can now redraw their own recent rows in a bounded live region:
  • Supported: CSI A/B/E/F, ESC M/D/E, save/restore (ESC 7/8, CSI s/u), CSI J 0/1 and CSI K 0/1/2.
  • The region never reaches above the newest NMSh-written line (command header, notice, earlier command) or more than the PTY's height (OutputBuffer.setScreenRows).
  • A command resumed from a journal is fenced to its own rows.
• Absolute row positioning is not mapped and keeps today's behavior. Copy and search see the final rows, as a terminal would show them.

Tests

Each new test fails on the unfixed code and passes with the fix:

• shellAdapters: a prepared Fish home starts no generator. The negative control finds the generator.
• shellIntelligence: a superseded lookup is answered from the new knowledge. queuedHighlight: a stored command whose lookup was superseded still ends classified (real SemanticService). Old code: never recolored, and settled unknown.
• ansiLiveRegion: 10 tests, including the Inline multi-line ANSI progress redraws can accumulate duplicate transcript lines #388 fixture through a real zsh PTY, interleaved log lines, fences, split sequences, resume, and 10,000 frames in about 35 ms. The old parser fails 9 of 10; the tenth is the documented unchanged absolute-positioning case.

Further runs:

• Flaky: queuedHighlight composer/queue classification is not settled after one event-loop turn #408 tests: 16/16 passes in two concurrent loops.
• All 59 test files touching the parser, OutputBuffer, the presenter or copy payloads: green.
• Full npm run verify locally:
  • The first run used a too-long temp root on the external drive. That failed the runner's own short-root check plus the tmux and socket tests, for environmental reasons only.
  • Re-run with a short root, those pass. The remaining live-PTY timeouts under full-suite load on this 8 GB Mac pass when run alone.
  • CI is the exact-head gate.

Not verified

Physical terminal checks are listed in docs/testing/v021-physical-qa.md (section 1) and have not been run.

Refs #403, #408, #388, #410