fix(release): retry gh pr merge to cover GraphQL propagation lag

[JohnnyVicious]

Summary

Wraps gh pr merge in the release workflow with a 5-attempt retry loop (2s backoff) to ride out GitHub's GraphQL propagation lag between gh pr create and the subsequent merge.

Why

The v1.0.4 release hit this in anger. Workflow run 24312750234 failed at step 11 with:

Created release PR: https://github.com/.../pull/53
GraphQL: Could not resolve to a PullRequest with the number of 53.
##[error]Process completed with exit code 1.

gh pr create returned synchronously with the new PR URL, but gh pr merge fired ~500ms later and the PR node wasn't yet resolvable from the merge endpoint — classic eventual consistency. I confirmed it was a pure race by checking PR #53 manually a few seconds later: it was already MERGEABLE, and gh pr merge 53 --squash --delete-branch worked first-try.

v1.0.2 and v1.0.3 releases happened to land on the fast side of the race; this fix is about hardening the automation, not reacting to a new failure mode.

Test plan

• YAML syntax parses (.github/workflows/release.yml is a single step edit)
• Retry loop logic is standard bash — attempts 1..5 with sleep 2 between, exits non-zero with a ::error:: annotation on final failure
• Real validation: the next release cut will exercise the retry branch if GitHub lags, or the happy path if it doesn't. Either way it ships v1.0.5+ cleanly.