Pick the tests to run for a diff

Your full regression suite takes 4 hours, so an agent picks the tests each diff needs. A missed regression in production is the nightmare, so whenever the agent is unsure, it runs everything.

Write select_tests_for_diff(test_records, changed_files): the rule your agent follows for each diff.

• test_records (list of str): the coverage map, one test each, written name|tag|files, where files is a list of the files the test covers separated by ;. For example "test_auth|unit|src/auth.py;src/util.py".
• changed_files (list of str): the files the diff touches.
• Return the names of the tests to run, each once, sorted in ascending order.

Rules:

1. Split each record on | into the name, the tag and the files. A missing tag or file list counts as empty. Split the file list on ; and drop empty entries.
2. Tests tagged exactly smoke or critical always run.
3. If no files changed: run only the always-run tests.
4. If any changed file is not covered by any test: run every test.
5. Otherwise: run the always-run tests plus every test that covers at least one changed file.