feat(prview): PR review-comment threading, diff hunks, and web-parity activity (closes #852)

[ysrfz]

Summary

• Closes issue PR activity view flattens review threads and drops the code context #852
• I have read the CONTIBUTING.md and AI_POLICY.md guides

Status: WIP — please comment if you don't like the current direction while I polish the edges.

Reworks the PR activity panel to look more like GitHub web's Conversation tab. A few things going on here:

• Thread grouping — review comments now sit under the thread they belong to instead of flat-by-timestamp. Replies get a compact ↳ {author} prefix.
• Diff hunks — fetch diffHunk on review comments and render the last 4 content lines above each thread (mirroring the collapsed preview on web). @@ headers get stripped since they're just positional metadata.
• Thread badges — [Outdated] and [Resolved] render inline next to the anchor, and wrap to the next line when the anchor is too wide rather than getting clipped.
• DISMISSED glyph — was rendering as blank space before. Now uses a faint ↺ so dismissed approvals don't look like rendering bugs.
• Implicit-Review unwrap — GitHub auto-generates a state=COMMENTED, body="" wrapper whenever a thread starts outside a formal review. The TUI was rendering these as empty xyz reviewed blocks. Now: 0 threads → hide; 1 → promote the thread up a level; 2+ → promote them all as siblings.
• Conversation count — the N comments figure in the title now matches the web (N) badge, including the slightly weird case where reviews with state=COMMENTED + body + zero inline comments get excluded (web treats them as redundant with plain IssueComments, even though it still renders the card — I went back and forth on this one, ended up matching web for consistency).

Keybindings and overall panel layout unchanged.

How Did You Test this Change?

• go test ./internal/tui/components/prview ./internal/data — added 14 new tests in activity_test.go covering each behaviour above (thread grouping, diff-hunk truncation, badges, DISMISSED glyph, implicit-review unwrap cases, count-badge web-parity)
• go build && go vet ./... clean
• Manual side-by-side against gh-dash v4.24.1 on a 20+-review PR (mixed thread/IssueComment/DISMISSED activity); will attach screenshots once the UI is polished

Images/Videos

(will attach side-by-side screenshots once I'm happy with the UI)

AI Usage Disclosure

Per AI_POLICY.md: wrote the code with Claude Code as a pair-programming assistant, with substantial AI involvement on the implementation drafts and tests. Each design decision (web-parity counting, implicit-review unwrap cases, diff-hunk truncation length, badge wrap behaviour) was driven and reviewed by me, with multiple rounds of pushback on the approach before landing. I understand the changes and can explain how the activity-panel pipeline interacts with the GraphQL fragment and the existing PR view — happy to walk through any part on request.

Closes #852

[dlvhdr]

Hey @ysrfz :)
I've updated my AI policy to disallow any AI usage.
I am willing to explore your idea though and I think a first good step would be to open an issue to discuss it.
We can then make sure to take performance into account as well as build a figma mock we can agree on.
wdyt?