init: skip redundant tracked checkout

[junior-ricon]

Summary

modules init no longer runs git checkout <track> when a tracked module clone is already on that branch.

That avoids firing module-local checkout hooks unnecessarily during routine tracked-module refreshes. In the observed fold case, a no-op checkout of the den module fired den's notes post-checkout hook and produced macOS cut/grep illegal-byte-sequence warnings from locked notes state.

Changes

• Skip checkout when current_branch == track.
• Preserve existing behavior for detached clones and clones on another branch.
• Add a BATS regression that installs a post-checkout sentinel hook in an already-current tracked clone and verifies modules init does not fire it.
• Regenerate README test count from 135 to 136.

Validation

• mise run test — 136/136
• mise run test test/init.bats — 15/15
• Patched direct smoke against Junior fold checkout: no illegal-byte warnings
• codebase lint "$PWD"
• mise exec -C "$PWD" shiv:readme@0.3.1 -- readme build --check
• git diff --check
• git pre-commit
• git pre-push — warning only: fresh branch had no upstream and inherited GitHub merge commits are not all locally good-signed; Junior commit 1816f5c is good-signed

[quick-ricon]

Adversarial review pass: approved.

Findings: none.

What I tried to break:

• Read the changed sync_tracked_branch path as shared architecture, not only init: this also affects tracked-module modules update no-op checkouts because the helper is shared. That looks like a desirable reduction in redundant checkout hooks, and existing update behavior still passes.
• Checked the preserved normalization paths: detached tracked clones still go through checkout, missing local branches still create a tracking branch, wrong current branches still checkout the target branch, upstream is still set after the skip, and remote-ahead/local-ahead/diverged logic still happens after the checkout gate.
• Verified the new regression is testing the real old failure mode: when I temporarily restored the old unconditional same-branch checkout, init skips checkout hook when tracked clone is already on target branch failed because the sentinel post-checkout hook fired.
• Ran targeted and broader validation locally in a correctly named temp worktree:
  • mise run test test/init.bats — 15/15
  • mise run test test/update.bats — 15/15
  • mise run test — 136/136
  • git diff --check origin/main...HEAD
  • codebase lint "$PWD" — all 9 rules passed when the worktree basename was modules (a first run in a differently named temp worktree produced only the expected caller-pwd-contract false positive from the package-name heuristic).

Surviving concerns / non-blocking notes:

• The PR title/body emphasize modules init, but the implementation intentionally changes the shared tracked-branch helper used by update too. I think that is fine, just worth remembering when reading the scope.
• Untracked-module update still has its own checkout behavior; that is outside this tracked-module no-op fix.

The design looks sound to me: skip only when already on the exact target branch, while retaining every path that needs checkout to restore branch shape or create tracking state.