feat(replay): group meaningful source-backed changes

[fcoury-oai]

Why

Raw Git hunks are not always meaningful review steps. A standalone blank-line hunk can inherit the name of a neighboring function, while a larger compatibility change can be reduced to its first added field and receive an unrelated pull-request summary. That makes PR Replay harder to trust precisely when reviewers are trying to understand what the original code actually does.

What changed

• Add an immutable-hunk-backed semantic presentation overlay that groups related consecutive changes within the same file without discarding their original diffs, stable identities, source anchors, or prerequisite ordering.
• Attach incidental whitespace changes to the nearest substantive change and combine derived-deserializer removal with the matching backwards-compatible deserialization implementation.
• Generate change titles, explanations, and bounded supporting details from actual added/removed source, Rust syntax ownership, compatibility behavior, and relevant author-written rationale instead of neighboring hunk headers or unrelated PR text.
• Apply, manually validate, undo, and recover every original hunk in a semantic group as one attributable editor operation while preserving individual original-hunk completion and inline-comment anchors.
• Render a bounded summary of important secondary behavior beneath the selected change and document the semantic overlay and original-hunk invariants.
• Cover whitespace-only hunks, unrelated neighboring symbols, backwards-compatible deserialization, cross-file isolation, duplicate/malformed groups, manual reconstruction, atomic apply/undo, crash recovery, and compact panel details.

How to Test

1. From a repository with a pull request containing a formatting-only hunk followed by a real change, launch this branch and start Replay with Space R g. Confirm the blank line is not presented as an independent named change, the substantive change remains accurately named, and its original diff still includes both exact hunks.
2. Review a change that removes derived deserialization, adds a field, introduces an explicit deserializer, and supplies a backwards-compatible default. Confirm Replay presents the full compatibility behavior with a relevant WHY and bounded supporting details rather than describing only the first field.
3. Select a grouped change, press a, and confirm every displayed original hunk applies together. Press u and confirm the complete group reverts in one undo without overwriting reviewer-authored changes.
4. Reconstruct the same grouped change manually and validate it; confirm all underlying original hunks are marked complete while the change list records one semantic completion.
5. Apply a grouped change, restart with --resume, and confirm its progress, original source anchors, and single-step undo are restored.
6. Run cargo test --lib and cargo clippy --all-targets --all-features -- -D warnings.