fix: use git blame -C to prevent misattribution of shifted lines in git ai diff

[svarlamov]

Summary

• blame_hunks_for_ranges in src/commands/blame.rs now translates detect_copies/detect_moves on GitAiBlameOptions into -C/-M flags on the git-blame command line.
• apply_blame_for_side in src/commands/diff.rs already set detect_copies: 1, but the field was never wired through to the actual git invocation — the fix was inert (caught by Devin review).
• Adds a regression test test_diff_blame_uses_detect_copies_for_moved_ai_lines that fails on the broken code and passes with the fix.

Root cause

When a commit moves a substantial block of lines to a different position within a file (e.g. reordering functions), Myers diff represents the moved block as explicit + lines. Those lines end up in apply_blame_for_side's added_lines set. Without -C (which implies -M), git blame attributes them to the current commit rather than the original one — so they fall through as Human when the current commit has no attestation covering them. With -C, git blame detects the within-file move and correctly traces the lines back to the originating commit.

Observed on commit 7b99854f (100% AI-authored): git ai diff showed 4 lines attributed to Human because git blame without -C placed them in the current commit rather than parent 6ff95531 where the covering attestation lived.

Regression test

test_diff_blame_uses_detect_copies_for_moved_ai_lines:

1. Commit A — AI writes func_one (6 lines) + func_two (6 lines), both fully AI-attested.
2. Commit B — AI adds new_func (attested in B) and func_one shifts to the end. Myers diff represents func_one as + lines at its new position.
3. Without -C: git blame says the moved func_one lines were introduced in B → no attestation in B → Human ❌
4. With -C: git blame traces them back to A → A's attestation covers them → AI ✓

Test plan

• cargo test --test integration -- diff — all 201 tests pass (1 new test added)
• The failing ubuntu wrapper-daemon CI check is test_search_by_commit_range_in_worktree — a daemon sync timeout unrelated to this change; the same test passes on ubuntu wrapper and ubuntu daemon modes.