Fix review reruns after thread responses

[piekstra]

Summary

cr review could exit as already approved or already complete after cr respond posted newer inline replies. GitHub records those replies as COMMENTED review events, so the merge gate correctly required a newer verdict while the normal review command refused to produce one.

This makes both live-review fast paths select the latest approval, change request, or comment from the posting identity. A newer comment now starts a fresh review; a latest approval or change request keeps the existing behavior.

Testing

• go test ./internal/gateio ./internal/app ./internal/cmd/reviewcmd
• golangci-lint run
• Built the patched CLI and used a normal review to produce a new exact-head approval after three thread-response events: dogfood review
• make check passes through the changed packages; the local installed-Pi compatibility test fails because that Pi version rejects --no-builtin-tools and --no-approve