Trace: batch staged blob reads during Bash commit checkpoints #3749

Problem

In #3631, checkpointActiveShellAttribution calls ExecClient.IndexFile for each changed staged path. Every call scans the full index with git ls-files --stage and starts git cat-file --batch for one blob, although afterIndex[relPath] already contains its hash.

In @jiparis's review, a Bash call that staged 301 files in a 3.3k-file checkout took 10.6 seconds at commit-msg; the per-file index/blob reads account for about 5 seconds. The reviewer explicitly accepted this as a follow-up, not a blocker for #3631.

Scope

Read only the changed staged blobs by their already captured hashes in one git cat-file --batch call. Preserve the effective-index snapshot (including split, sparse, and temporary indexes), the GoGit fallback, exact binary content, and existing same-call attribution behavior.

Acceptance criteria

• The checkpoint takes one index snapshot and one batch blob read, without a full-index scan or Git process per changed file.
• Regression tests cover multiple staged files, binary content, and hashes captured before a later index change; existing Bash commit lifecycle tests pass.
• Record comparable before/after timing for the many-file scenario in fix(trace): attribute commits made inside Bash calls #3631, without a timing-sensitive test assertion.

Implemented in #3631 by commit 77d2fe0e; the many-file timing remains to be recorded.