fix(snapshot): seed index from the worktree's git dir

[Hyyy6]

Issue for this PR

Closes#39388

Type of change

• Bug fix

What does this PR do?

#31798 seeds the snapshot repo with the source repo's git index so git add --all
reuses already-hashed entries instead of re-hashing the whole tree. In a linked worktree
--git-common-dir resolves to the main repository's .git, so the copied index is the
main repo's, which does not match the worktree's working tree. git add --all then
re-hashes every file, reintroducing the hang on large worktrees.

Resolve the index from --git-dir (cwd = the worktree) instead, which in a linked worktree
points at the worktree's own .git/worktrees/<name> (whose index matches its working tree).
Falls back to the common dir's index when rev-parse fails, preserving the existing
best-effort behavior.

How did you verify your code works?

End-to-end (opencode run "say hi" in a chromium --depth 1 worktree, ~500k files):

Binary	Location	Time	Result
Official 1.18.4	main repo	9.8s	completes
Official 1.18.4	worktree	1m50s+ (killed)	hangs — git add --all re-hashing
This fix	worktree	38.5s	completes, model responds

Isolated git-level reproduction (replicating Snapshot.seed() at packages/opencode/src/snapshot/index.ts:198-239):

Scenario	Index source	Time	Tree hash
A: no seeding (pre-#31798)	none	>180s (timeout)	4b825dc6... (empty)
B: common-dir index (bug)	main repo	166s	36e0ef8a...
C: git-dir index (fix)	worktree	6s	36e0ef8a...

B and C produce identical tree hashes — the fix preserves correctness. C is 27x faster.

• bun test test/snapshot/snapshot.test.ts from packages/opencode — 55 pass, 1 skip, 0 fail.
• bun typecheck from packages/opencode — clean.

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[Hyyy6]

test, typecheck, and nix-eval are queued behind the first-time fork contributor approval gate. Everything else is green. I've already run bun typecheck and bun test test/snapshot/snapshot.test.ts locally in packages/opencode (55 pass, 1 skip, 0 fail) — happy to re-run anything once workflows are approved.

[Hyyy6]

@nexxeln — you self-assigned #39388; this PR is the fix. It's a ~3-line change, reproduced and verified on a ~500k-file Chromium worktree as in original related PR. Happy to rebase onto current dev if needed.