perf(ios): reduce capture probes and text entry waits
Summary
Reduce iOS latency by overlapping fresh system-host scans, deferring irrelevant keyboard reads, and shortening unreadable-value warmup. Retain first-character target checks and final verification.
On the upstream test app, median warm captures improved from 91.5 to 75.8 ms and password replacements from 3,317 to 3,044 ms. Removing the split was rejected after it typed into a successor input.
The further two-second fill investigation rejected Select All clearing: a provisional 2.55-second sample lacked login verification, and a masked-input regression failed while unchanged clearing passed. No further runtime shortcut is retained.
Scope: six runtime/test files and four evidence files. Context: tester-army/e2e#1236. Measurements, rejected experiments, and reproduction.
Validation
Tested commit: 8a3931a9c4fd8188636b7f5a36d5e56b6b97b038; comparison base: 8f3eb3d2efde3654a560fddb39127d556e068352.
pnpm check:affected --run --head 8a3931a9c4fd8188636b7f5a36d5e56b6b97b038: passed. Receipt.tmp/validation/run-wL2vPV/receipt.json:exactCommit=true,checkoutChanged=false; nine checks passed, two deduplicated, six skipped. 1,046 related tests passed.- Runtime verification: 17 native regressions and 43 focused capture/routing tests passed; iOS/macOS builds passed. Follow-up: control 3 passed; rejected candidate 2 passed/1 failed.
- Independent review: no remaining findings.
GitHub owns coverage, native lanes, and macOS replay; manual iOS replay lanes are parked. Physical-device verification and the full upstream suite were not rerun.