fix: bound Cursor Connect frame buffering

[Ingwannu]

Summary

• keep uint32-compatible outbound Connect frames while limiting inbound payloads to 32 MiB
• reject an oversized declared length as soon as the five-byte header arrives
• incrementally complete one pending frame instead of repeatedly concatenating it with unrelated following frames
• fail a turn when EOF leaves an incomplete Connect frame

Part of #820. The 32 MiB inbound limit matches the repository existing framing-hardening design note (devlog/_fin/350_cursor-provider-add/116_phase33-connect-framing-robustness-plan.md).

Verification

• bun test tests/cursor-framing.test.ts tests/cursor-hardening.test.ts tests/cursor-live-transport.test.ts (45 pass)
• bun run typecheck
• bun run privacy:scan

Compatibility

Normal split headers, split payloads, and multiple frames per chunk preserve their order. A legitimate inbound frame larger than 32 MiB now fails explicitly; outbound framing remains unchanged.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 7ce4af2c-7ef7-4ec9-a45e-8f2170610a4c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[Ingwannu]

CI follow-up: the initial macOS failure was the same Bun 1.3.14 Trace/BPT trap / bun.report signature seen on unrelated PR #841. The failed job passed on rerun, while the other platform jobs were already green. No code change was needed.

[lidge-jun]

Maintainer triage: this is a narrow and valuable hardening change. It bounds inbound Cursor Connect frame buffering and rejects incomplete streams instead of allowing unbounded or ambiguous framing state.

The branch currently conflicts with dev, so it needs a rebase before it can land. It also shares the Cursor transport/docs surface with #845; please land the two PRs in a defined order (or consolidate them) to avoid overlapping changes and documentation drift.

Because this is co-maintainer-authored, an independent non-author review is still required before landing. Thanks for keeping the scope focused.

[lidge-jun]

Superseded by #892, which implements this defect's fix natively on top of the wave-1 retained-state framework (the PR's shape was used as a reference; see the PR description in #892 for the defect-specific deltas, including where the framework already covered part of it). Thank you for the contribution — the analysis in this PR drove the fix.