[Rahmat Eka Satria (rhmatzeka)]

Fixes #25742. This continues #25743 by Andrew Lavin (@andravin), who said they no longer have time for it; their commits are credited as co-author.

Tests hidden by a Test Explorer filter (anything in TestRunRequest.exclude) were still started and run. This builds the excluded set once, including descendants, and skips those items when expanding the request into test cases, in both the workspace path (WorkspaceTestAdapter.executeTests) and the project-based path (executeTestsForProjects).

The review comments on #25743 are covered:

• Rebased on main. It's a fresh branch off current main, which now has the project-based execution path.
• Test-ID deduplication is preserved. Both paths still collect run IDs through the existing sets; exclusion only filters nodes before that. There's a test for an overlapping selection ([TestClass, test1] with test2 excluded) sending test1 once.
• pytest is not invoked when exclusions leave no tests. The workspace path now returns before calling the execution adapter when an exclude set was given and no IDs are left, since an empty ID list makes pytest run everything. The project path already returned early on an empty list; there's a test that a project whose tests are all excluded is skipped while the other project still runs.

One unrelated test change: the existing "the right tests should be sent to be executed" test stubbed getTestCaseNodes with sinon.stub instead of the suite's sandbox, so the stub was never restored and leaked into later tests in the suite. It now uses sandbox.stub.

Testing: npm run test:unittests passes for everything under src/test/testing/testController (276 tests); 12 of the new tests fail against the old code. The only failures in the full suite are environment-related on my machine (no python on PATH, no native locator binary) and are in files this PR doesn't touch. eslint and prettier --check pass on the changed files.

[Rahmat Eka Satria (rhmatzeka)]

@microsoft-github-policy-service agree