BE-1019: Replace Travis CI with GitHub Actions

[daniel1991gomes]

Migrates CI from Travis to GitHub Actions per BE-1019. OTP matrix modernized to 24.3/25.3/26.2 (Travis pinned 21.0–22.1, which predate current GitHub-hosted runners); adjust if the team targets other versions. Adding the workflow means it runs on this PR.

Automated triage, draft, needs human review.

Closes BE-1019 · https://airtable.com/appf4OQE23a6VJfC0/tblmshSXKQfEHeWWf/recjyR6MHNP4sGyLB

[daniel1991gomes]

CI is red here, but the failures are pre-existing and unrelated to this CI migration (which only swaps Travis for GitHub Actions), so they're intentionally not fixed in this PR.

The tests were last touched in 2018–2020, when Travis ran them on OTP 21/22. They haven't changed since; OTP has. Two failures, both consistent with that drift:

1. statman_decorators:no_key_test/0 — asserts the exact internal name the compiler gives a decorator-wrapped function, and that naming changed in newer OTP ('-no_key_decorator1___/2-fun-0-' → '-fun.no_key_original___/2-'). OTP-version-specific.
2. statman_histogram setup: ets:new(statman_histograms, [named_table,...]) → badarg (already_exists). That global named table is created by fixtures in both test/statman_tests.erl and src/statman_histogram.erl; in a single-VM rebar3 eunit run it leaks across modules when a test doesn't clean up. A test-isolation fragility.

Both should be handled as a separate task. The migration itself (remove .travis.yml; add a workflow that compiles + runs the suite) is unchanged in scope.

(Correcting an earlier version of this comment that claimed the failures were "OTP-agnostic" — that was wrong; failure #1 is specifically an OTP-version change.)