fix(web): review pane keeps scroll on refresh and comments are editable

[kalvenschraut]

Problem

Refreshing the review pane on a large diff threw the reader back to the top. Large diffs load file by file, and that path keyed its state on the diff's content hash, so every refresh that changed the diff remounted the viewer, reset the loaded-file window to four files, and started every file patch from scratch. Review comments also could not be edited once saved, in the review pane or in a pull request review.

Fix

• useReviewFilePatches scopes its state to the comparison (environment, cwd, source kind, base ref, whitespace) instead of the content hash. A refresh refetches the loaded files in place. Effect atoms keep the previous result while waiting, so the viewer reconciles items by their stable ids and Pierre's scroll anchoring holds the position. Switching scope, turn, base ref, or whitespace still resets the view on purpose.
• DiffCommentAnnotation gets an edit mode for saved comments (pencil button; Save or ⌘/Ctrl+Enter, Esc to cancel). It is wired into the review pane and file preview comments. addReviewComment now replaces an edited comment where it stands instead of moving it to the end.
• Pending pull request review comments get the same edit button, using the existing PullRequestMarkdownEditor and a new updateComment store action. After a successful submit, removeComments removes only comments whose body still matches what was sent, so a comment rewritten while the request was in flight stays pending.

Verification

• vp test run for composerDraftStore, DiffCommentAnnotation, and all of components/pullRequest (505 tests), including new tests for in-place edits and the in-flight edit race.
• Web typecheck and lint on the changed files are clean.
• Reviewed by GPT 6.1 Sol: round 1 found the in-flight submit race (fixed here), round 2 approved with no findings.
• Not exercised in a browser.

Done by Claude Opus 5.5 in Claude Code (T3 Code), reviewed by GPT 6.1 Sol.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 952e8b8 · PR result: b80dc19 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.