fix: correct compaction token estimates and completion state

[SudoJacky]

Compaction could report identical before/after token counts despite replacing the history, and a completed manual compaction could leave the chat showing a running response.

This change estimates the replacement provider request after clearing stale Responses replay on a projected context copy, matching the actual checkpoint installation. Ordinary requests retain their native replay. After manual compaction, the chat reloads the persisted terminal state; pending streaming frames cannot overwrite that state, and new live updates invalidate older in-flight loads.

Regression coverage checks compaction and trimming estimates against the resumed request, cached running state after compaction, and both timeline reload/frame ordering cases. Required architecture and module documentation are updated.

Validation completed locally during implementation:

• 35 Rust context tests passed.
• Related frontend runtime, service, projection, and usage tests passed.
• TypeScript type checking, Rust formatting, and diff checks passed.
• Commit hooks passed engineering documentation and module README freshness checks.

Scope is limited to the compaction fixes, regression tests, and their required technical documentation.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-20T04:43:21.310029Z	811ac81	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.