[imfing]

Summary

• diffs branch BASE HEAD reviews HEAD against BASE (same as git diff BASE...HEAD) without either branch being checked out; omitting HEAD keeps today's behavior
• git::branch_diff takes an optional head ref; --include-dirty conflicts with it (clap) and is a 400 from /api/branch-diff, since uncommitted changes only make sense for the checked-out branch
• An unknown head ref fails fast in the CLI (unknown head ref: <name>)
• /api/blob accepts rev=<ref> with path, so pure renames (no blob oids) load from the reviewed branch instead of the working tree
• Local comments accept an optional ?branch= scope, so threads on a non-checked-out branch are filed under it; diffs comments is unchanged (None scope)
• Web forwards ?head=, shows it in the title/branch chip/export, and keeps it when switching base

Much of the comments.rs diff is cargo fmt re-wrapping existing call sites that now take a scope argument.

Test plan

• cargo test, cargo clippy --all-targets -D warnings, cargo fmt
• tsc, oxlint, oxfmt, vite build
• New tests: parity of branch_diff(main, Some(feature)) vs git diff main...feature with a third dirty branch checked out; DirtyWithHead error; read_rev_file; explicit comment scope; comment scope query validation; CLI target/label encoding
• End-to-end smoke test on a scratch repo (checked out on a dirty third branch): served diff contains only feature's edit + rename, renamed file loads from feature, head+dirty returns 400, comment filed under feature