Run API compliance tests automatically for API-touching PRs

[enyst]

Problem

We have API compliance tests, but they are currently optional/manual:

• label-triggered or manually dispatched
• not a default enforcement path for API-touching PRs
• configured with continue-on-error

So today they are a useful tool, but not a reliable part of public API enforcement.

What we may want

Automatically run at least a reduced API compliance suite for PRs that likely affect public APIs, for example:

• PRs touching openhands-sdk/
• PRs touching openhands-agent-server/
• PRs labeled api-change
• release PRs

Possible rollout options:

1. Run a smoke subset automatically for likely API changes.
2. Keep the full suite manual/label-triggered.
3. Fail loudly or at least post a very visible result summary.

Why

Structural diff checks are necessary, but runtime/API compliance coverage gives extra confidence that behavior remains usable for real clients.