test: delete low-value tests and dead test-only exports#24058

Four restore PRs (#23977, #23980, #23991, #24011) bring back 29 test cases the October snap over-deleted. The deal was: delete at least twice as many low-value tests first, so the suite does not just grow back. This is that deletion PR, and it should merge before the restores.

It removes 126 test cases: 1,503 net test lines, 1,050 net source lines, 2,561 net lines total. That is 4.3x the restored case count and 1.9x the restored test lines.

What went, by class:

• Dead exports with their tests: ACP runtime event builders, Grok background task events, compareDateTimeStrings, findSharedSettingsMismatches, legacyLinkedPullRequestOf, work-log command detail helpers, resolveProviderForCwd, beginCodexFeedbackSubmission, and two dead mobile thread-activity helpers (353 source lines). Every symbol was grep-verified caller-less across the repo before deletion. Dead package.json subpath exports went with them.
• Duplicates with stronger twins: checkpoint prefix coverage stays in the review-diff tests plus the restored checkpoint case (see review note), stdin ordering stays in processRunner.test.ts, update reducer cases stay in DesktopUpdates.test.ts, the web threadSort shim test duplicates client-runtime.
• Copy and schema mirrors: MCP tool description strings, orchestration sections joined from their own constants, decode-then-read-back contract cases.
• Trivial table rows: shell-program samples, keyboard label variants, amount spellings, timestamp rows. Every implementation branch keeps at least one row.

Review caught 9 cuts that went too deep, and they are restored in the second commit: the checkpoint diff.noprefix guard (#9438), the preview-tools allOf guard (#5128) and tabId check, the troubleshooting doc-anchor test, one threadAccess wiring case (#22452), four timestamp rows (#20094), three amount-regex branches (#21519), the release-notes heading skip (#9138), the literal Ctrl++ row, and both multi-model send guard rows.

Checked: vp run check:changed passes (format, lint, knip, graphql, typecheck 32 packages, 404 tests in 30 files). All surviving touched test files were run directly.

Reviewed with sol-loop: 2 rounds with GPT-6.1 Sol on high. Sol verdict: approved. Also reviewed by Claude Opus 5.5 over 2 rounds: round 1 found the 9 over-cuts above, round 2 approved. Two nits were declined with reasons in review (a circular sections test deletion, minor updateMachine assert loss).

Created with Muse Wonder Full in Muse Code. Research and implementation by Muse Wonder Full subagents.