`gh stack rebase` uses the wrong onto boundary for merged branches

Start in internal/github/github.go and cmd/utils.go, then read cmd/rebase_test.go for the existing rebase boundary behavior. Run the rebase test and review the linked commit for the intended remote-versus-local SHA scenario. Done means the test covers local SHA A versus remote SHA B and gh stack rebase selects the remote merge point.

Problem: gh-stack needs to know "what was branch X's last commit right before it got merged" so it can skip replaying that history into upstack branches. Currently, it resolves this from the local clone instead of GitHub.

So, if branch X gets rebased elsewhere before merging, the local copy goes stale. That's normal and expected, however gh stack rebase fails to notice this and tries to replay from the wrong commit. This has led to spurious merge conflicts on files I didn't even touch, leading me to rebasing branches one-by-one or fixing the gh-stack state by hand (or with an agent).

Fix: I wrote a fix here. Full disclosure: I used Claude to write the fix but I manually reviewed and edited it afterwards.

I would open a PR for this but it's locked to contributors only.

Changes:

1. internal/github/github.go: Add one field, HeadRefOid, to PullRequest. This tracks the remote's record of the exact commit which was merged.
2. cmd/utils.go:
   • When a PR is detected as merged, save its HeadRefOid value.
   • When computing the rebase boundary, use HeadRefOid instead of getting the (possibly stale) local SHA through git.
3. cmd/rebase_test.go: Add a test that fakes the problem scenario ("local has SHA A, but remote has SHA B (the real merge point)" and checks the tool picks B. This test fails without the above fixes, and passes with them.

Net effect: The onto boundary now comes from remote / GitHub (can't go stale) instead of local, eliminating confusing rebase conflicts due to a misconfigured rebase.