#4394 fix: review follow-ups (web text-entry hint, recorder echo names, mutation report on cancel)

Summary

This PR collects non-blocking review follow-ups from merged PRs. It touches 7 files. Refs #4368.

• Web text-entry hint (#4362): a refused coordinate fill runs keyboard type underneath, so its hint told the user to retry type. The hint now follows the user's command. fill, by ref or by coordinates, gets the snapshot/ref hint, and type gets "focus the field again". The coordinate-fill row in the test table now expects the fill hint. It failed before this change.
• Linux type pin (#4362): two runtime asserts restated object literals and could not fail, so they are removed. The @ts-expect-error pin stays as a compile-time proof.
• Recorder echo names (#4363): parameterizeRecordedResultEcho now returns the placeholders it inserted. The recorder no longer infers them by counting occurrences, and its three inference helpers are gone. A literal replaced inside an object key is now marked live too, which matches what the recorder wrote.
• repairStateMode doc (#4369): the doc now says it forces any pre-existing state path to the private mode.
• Mutation-affected report (from the #4356 investigation): the report job condition changes from always() to !cancelled(). A run cancelled by a newer push no longer scores partial shards as a harness failure. Failed shards are still reported.

Validation

• Tested SHA 9ba7240. pnpm check:quick passes. pnpm check:fallow --base origin/main passes, with one intended weakened-tests warning for the removed dead asserts.
• Focused runs pass: 1,304 unit tests across the touched packages, plus the recorded-fill tests (29) and the active-session-script-publication provider-integration test (8).
• pnpm check:affected --run (exactCommit): every step before integration-node passed. Integration-node failed only on the 2 smoke-daemon-clean tests, which fail in the container because PID 1 does not reap zombies. CI is authoritative for those, and for the full unit and coverage runs.
• The workflow change can be observed only on a real cancelled run.

---

Generated by Claude Code

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}