ci: build and test on push/PR

[chrisdebian]

publish.yml only builds/tests on a version-tag push (v*) or manual workflow_dispatch, so ordinary pushes and pull requests to main currently aren't built or tested at all. This adds a separate, additive ci.yml that runs on push (main) and pull_request, matrixed across ubuntu-latest and windows-latest since the Windows-only code paths (System.IO.Ports, System.Speech, PortAudio native runtime) are currently only exercised at publish time.

Doesn't touch publish.yml or its release behaviour.

Note on test flakiness: while verifying this locally, RotatorControllerTests.Update_runs_on_worker_thread_and_tracks_satellite failed once across three full-suite runs (passed 5/5 in isolation). Looks like a pre-existing race between the worker thread's completion signal and the position-snapshot refresh under full-suite parallel load, unrelated to this change — flagging so it's not a surprise if this workflow shows a failure here on an early run.

Happy to adjust scope (e.g. drop the Windows leg, or start Linux-only) if preferred.

[magicbug]

Thanks — happy to take a PR CI gate; a couple of scope tweaks please:

1. PRs only — drop the push / main trigger. I do not need CI on every push to main; gating pull requests is the useful part.
2. Linux only for now — start with ubuntu-latest only (no Windows matrix). Cheaper/faster check is enough for an initial PR gate; we can add Windows later if it earns its keep.

Something like:

on:
  pull_request:
    branches: [main]

jobs:
  build-and-test:
    runs-on: ubuntu-latest
    steps:
      # ... setup, build, test as you have them

Noted on the rotator test flake — unrelated to this PR, but good to know if CI goes red once or twice.

[magicbug]

Quick follow-up on the rotator flake you flagged (RotatorControllerTests.Update_runs_on_worker_thread_and_tracks_satellite): that is now fixed on main.

Root cause was UpdateSynchronously signalling completion before display Az/El were polled (filled on the next worker loop tick). Track path now polls immediately after SetPosition, and the test disposes the controller so the worker does not keep running under parallel suite load.

Should not be a surprise red once the PR CI workflow is in.

[chrisdebian]

Both changes made: dropped the push/main trigger (PR-only now), and dropped the Windows leg for this initial gate (ubuntu-latest only). Thanks for the quick review.