An issue whose body says `Relieves: <path>` is not held by that file being hot, one such build at a time per file #7431

An issue with a Relieves: <path> line is no longer held because that file is hot, but only one such build per file runs at a time. tsc is clean and the targeted tests pass. The new test, the hot-file tests and the map-holds test ran; I did not run the full suite (the forge runs it on push).

• Exemption: relievedFiles in src/crew/hot_files.ts reads the Relieves: line, uses the same path rules as Files:, and keeps only paths that are also on the Files: line, at most 3. hotOverlap leaves those paths out. An issue with no Files: line relieves nothing.
• One at a time: src/crew/held.ts tracks which paths an open pull or a run in flight is already relieving. A second relieving issue on a hot path stays under hot_file. The map detail and the feed line say waits: <path> is already being relieved by #N.
• Same pass: if two relieving issues are ready together, only the first starts.
• Pins: a pin from file_holds still holds the issue. The pin check uses expectedFiles, which Relieves: doesn't change.
• Docs: docs/PLANNING.md has a new section, "Relieving a hot file", with the two-sentence rule for planners.
• Feed wiring: holdHot now takes the relieving map as an optional last argument, and src/crew/autonomy.ts passes it, so the feed shows the same sentence.
• Tests: test/hot-files-relieves.test.ts covers the four cases from the issue, plus the 3-path cap and the same-pass rule.

Deliberately left out: no Relieves: line on #7299 itself. A planner or person adds it when they decide that issue's purpose is to cut that file.

Closes #7422

An issue with a Relieves: <path> line is no longer held because that file is hot, but only one such build per file runs at a time. tsc is clean and the targeted tests pass. The new test, the hot-file tests and the map-holds test ran; I did not run the full suite (the forge runs it on push).

• Exemption: relievedFiles in src/crew/hot_files.ts reads the Relieves: line, uses the same path rules as Files:, and keeps only paths that are also on the Files: line, at most 3. hotOverlap leaves those paths out. An issue with no Files: line relieves nothing.
• One at a time: src/crew/held.ts tracks which paths an open pull or a run in flight is already relieving. A second relieving issue on a hot path stays under hot_file. The map detail and the feed line say waits: <path> is already being relieved by #N.
• Same pass: if two relieving issues are ready together, only the first starts.
• Pins: a pin from file_holds still holds the issue. The pin check uses expectedFiles, which Relieves: doesn't change.
• Docs: docs/PLANNING.md has a new section, "Relieving a hot file", with the two-sentence rule for planners.
• Feed wiring: holdHot now takes the relieving map as an optional last argument, and src/crew/autonomy.ts passes it, so the feed shows the same sentence.
• Tests: test/hot-files-relieves.test.ts covers the four cases from the issue, plus the 3-path cap and the same-pass rule.

Deliberately left out: no Relieves: line on #7299 itself. A planner or person adds it when they decide that issue's purpose is to cut that file.

I covered the feed wiring with a test and documented the author question instead of enforcing it. tsc is clean and test/hot-files-relieves.test.ts passes.

• Feed wiring: the new test calls holdHot with a relieving map. It asserts the recorded hold in standingHot carries relieving and that the feed sentence reads #N waits: <path> is already being relieved by #M. The test calls holdHot directly, not driveRepo. Deleting the held.relieving argument in autonomy.ts would still leave it green. That argument is only covered by tsc.
• Author gate: I left Relieves: unchecked against the author, and docs/PLANNING.md now says that is deliberate. The reason is that the line lifts only the hot-file count. It admits at most three paths the issue also lists under Files:. Other holds, pins and the one-relieving-build-per-file limit still apply, so a stray Relieves: can't pile pulls up.
• Why not enforce: the exemption is evaluated in hotOverlap, mapHolds and holdHot. None of them has the author or the database. Gating them would mean threading trust through all three, which seemed more than this bounded risk justifies. If you still want it enforced, say so and I'll do it, with a guest-authored test.