chore: remove test-only dead exports

[thymikee]

Summary

Removes 13 production exports that were consumed only by their own unit tests, plus those tests. These are exports kept "alive" purely by a test that exercises them — no production code path (internal, re-export, registry, or dynamic dispatch) reaches them.

28 files changed, −1121/+31.

How they were found

A fallow production-vs-default differential:

• fallow dead-code --unused-exports --production (tests excluded) flagged 231 exports.
• fallow dead-code --unused-exports (tests count as consumers) flagged 2.
• The difference = exports used only by tests.

Each of the 229 was then reachability-traced and classified into dead / live-internal-seam / false-positive, and every "dead" verdict was adversarially re-verified by an independent pass that tried to find any production use. That filter matters — it overturned 4 candidates that looked test-only but were actually reachable (e.g. buildSnapshotVisibility, withAppleToolProvider). This PR ships only the 13 high-confidence, fully-verified removals.

Removed exports (+ their dedicated tests)

Export	File
commandNameSet	commands/cli-grammar/common.ts (no test — pure dead)
openApp (standalone fn)	daemon-client.ts — distinct from the live Backend.openApp method
resolveIosDevice	core/dispatch-resolve.ts
stopRunnerSession	platforms/ios/runner-session.ts
xctestrunReferencesExistingProducts	platforms/ios/runner-xctestrun-products.ts
capturePostGestureStabilizedSnapshot	daemon/post-gesture-stabilization.ts
annotateAndroidScrollableContentHints	platforms/android/scroll-hints.ts
parseReplayScript	replay/script.ts
listDaemonHandlerCommands	daemon/daemon-command-registry.ts
getOptionSpecForToken	utils/cli-option-schema.ts
findNodeByLocator	utils/finders.ts (not the public finders.ts entry)
assertBackendCapabilityAllowed	runtime.ts
MAESTRO_COMPAT_UNSUPPORTED_CAPABILITIES	compat/maestro/support-matrix.ts

Removing assertBackendCapabilityAllowed cascaded into now-unused backend.ts helpers (hasBackendCapability, hasBackendEscapeHatch, and the BACKEND_CAPABILITY_ESCAPE_HATCH_METHODS map), which are removed too.

Preserved coverage

Tests whose real subject is live code but which incidentally used a removed symbol were rewritten to keep their coverage, not deleted:

• daemon command-registry / request-handler-catalog routing tests (now assert via the live getDaemonCommandRoute / catalog).
• getRunnerSessionSnapshot status test (dropped only the stopRunnerSession teardown).
• writeReplayScript whitespace test (asserts serialized output directly instead of round-tripping through parseReplayScript).
• Maestro doc-sync test (iterates only the still-used supported-capabilities const).

Deliberately NOT included (follow-ups)

Three other test-only exports were found but left in place because removal isn't mechanical:

• parseArgs (utils/args.ts) — a test-only wrapper over the production-live parseRawArgs+finalizeParsedArgs; its tests cover live parse behavior and live in args.test.ts alongside the CLI-help assertions. Needs re-pointing, not deletion.
• listCapabilityCommands / listCapabilityCheckedCommandNames — test-only consistency guards asserting the capability matrix and parser schema stay in sync. The functions are test-only but the invariant is real.

There are also ~170 live-internal-seam exports (logic runs in production; export exists only for a unit test) that could be un-exported as a separate lower-risk sweep.

Validation

• pnpm typecheck ✓
• pnpm lint ✓
• pnpm test:unit ✓ (2568 tests; the 3 fillAndroid flakes are the known process-spawn contention flake in an untouched file — pass in isolation in 4.7s)
• fallow production unused-exports: 231 → 218, no new dead-everywhere exports introduced.

🤖 Generated with Claude Code

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.3 MB	1.3 MB	-5.5 kB
JS gzip	422.3 kB	420.2 kB	-2.1 kB
npm tarball	557.3 kB	555.6 kB	-1.7 kB
npm unpacked	1.9 MB	1.9 MB	-5.5 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.6 ms	27.7 ms	+0.1 ms
CLI --help	44.8 ms	45.0 ms	+0.2 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/5299.js	+72.3 kB	+23.3 kB
dist/src/9722.js	+995 B	+240 B
dist/src/cli.js	-613 B	-180 B
dist/src/record-trace-recording.js	-22 B	-2 B
dist/src/session.js	+1 B	+1 B

[thymikee]

parseReplayScript() was only a thin wrapper, but the parsing logic it exercised is still live through parseReplayScriptDetailed() and production callers in compat/replay-input.ts / compat/maestro/export-flow.ts. This PR deletes the record/screenshot/snapshot/gesture/type parsing assertions from this file instead of re-pointing them to parseReplayScriptDetailed(...).actions, so regressions in the .ad parser for flags like --hide-touches, --fullscreen, --force-full, gesture positionals, or --delay-ms would no longer be caught by unit coverage. Please keep those parser cases and call the detailed API directly, or add equivalent coverage through a live replay boundary.