fix(input): centralize and persist terminal input

[Lint111]

Summary

• Make TerminalInputController the single pre-delivery owner for native xterm
  data, Android helper-textarea mutations, IME composition, CJK/voice input,
  paste, accessory controls, and modified Enter.
• Persist editable drafts independently per session across session switches,
  page reloads, browser minimization/tab discard, and phone sleep without
  submitting restored text.
• Queue submitted input durably before transport, preserve ordering across
  WebSocket/HTTP reconnects, and remove frames only after the server confirms an
  accepted PTY write.
• Extend xterm-zerolag-input at its source package with serializable,
  grapheme-aware multiline state and atomic restoration against stale terminal
  frames.

User-visible fixes

• Prevent delayed Android/Gboard composition replays such as cdcd and
  homecd.
• Keep Backspace at one action per character instead of consuming an invisible
  autocomplete/helper-textarea buffer first.
• Preserve multiline and blank-line paste data, including bracketed-paste
  framing for large payloads.
• Keep unfinished input attached to the correct session and editable after
  lifecycle or session transitions.
• Preserve drafts while permission/elicitation dialogs use navigation and
  selection controls.
• Keep local echo, cursor placement, and multiline viewport positioning
  coherent while terminal output arrives.

Delivery model

Each browser has a stable clientId; each session gets monotonic input sequence
numbers. Pending frames are stored before delivery and retried after reconnect.
The WebSocket route ACKs tagged input only after the PTY accepts it, while the
HTTP route uses a successful response as its ACK. Duplicate (clientId, seq)
frames are acknowledged without a second write.

Deduplication is intentionally server-memory-only. A server restart in the
narrow interval between a successful write and client ACK can produce a retry;
the design favors retaining user input over losing it.

Review scope

This is the IME, paste, draft, and input-delivery slice requested in the review
of #173. It does not include the mobile navigation pad, shutdown endpoint,
repository viewer, response viewer, history streaming, hook changes, or case
actions; those are separate PRs.

The diff is 45 files and +7,283/-1,006. Of the added lines, 4,032 (55%) are
focused tests. Generated dist/ and vendor bundles are not committed; the
zero-lag overlay remains sourced from packages/xterm-zerolag-input/.

Commit structure

1. feat(zerolag): add serializable multiline input state
2. feat(input): add per-session terminal draft store
3. refactor(input): centralize terminal input arbitration
4. feat(input): integrate persistent terminal drafts
5. refactor(input): route producers through controller
6. test(mobile): cover persistent terminal input
7. fix(input): acknowledge only accepted terminal writes
8. docs(input): document persistent delivery invariants

Validation

• Input/controller/transport suites: 12 files, 223 tests passed
• Mobile keyboard browser suite: 1 file, 57 tests passed
• xterm-zerolag-input: 5 files, 194 tests passed; typecheck and package build
  passed
• Root typecheck, ESLint, Prettier, frontend syntax (28 files), and public asset
  checks (45 files): passed
• Production build, including minified/content-hashed input state and controller
  assets: passed
• Full CI-equivalent run on the audited combined stack with the isolated test
  harness from test: isolate runtime state and PTY integration #175: 190 files passed, 1 skipped; 3,860 tests passed, 12 skipped

Related

• Extracted from Improve mobile terminal reliability and session tooling #173
• Test isolation: test: isolate runtime state and PTY integration #175
• Mobile terminal tap behavior: fix(mobile): route terminal taps without opening the keyboard #186

[Ark0N]

Hi Lior, closing this one as part of a cleanup of the seventeen PRs currently open from you. The full explanation is in #173, and I would rather you read that one than this note, because it is the honest version and it is not a dismissal of your work.

The short form: sixteen PRs opened in a single day, roughly 42,000 added lines in total, landing mostly on the same few files (nine touch app.js, eight touch terminal-ui.js), with five already conflicting against master and none carrying CI results. I cannot review that at the standard this codebase needs, and merging it at a lower standard would be worse than not merging it.

Going forward, please keep no more than three open PRs at a time, each one a single behavior change that stands on its own, with a test that fails on master and passes with the fix. #214 and #215, merged today, are good models.

If this particular change fixes something that genuinely annoys you in daily use, it is a good candidate to be the first one you reopen on its own, rebased on current master. I will review it properly.

Thanks for the effort you put in, and sorry to close it this way.