test_runner: add skip, todo, only, and expectFailure to subtest context

[Felipeness]

The top-level test() function exposes test.skip(), test.todo(), test.only(), and test.expectFailure() variants, but these were missing from TestContext's test() method. Calling t.test.skip() inside a test callback threw TypeError: t.test.skip is not a function.

The documentation states that t.test() "behaves identically to the top level test()", so this is a bug.

Changes

• Move test() from the TestContext class prototype to an arrow function in the constructor, allowing variant methods to be attached as properties
• Extract a shared runSubtest() helper to avoid duplicating plan counting and createSubtest logic
• Attach .skip, .todo, .only, and .expectFailure — matching the exact same list used in harness.js for the top-level test()

Trade-off

This trades one shared prototype method for 5 closures per TestContext instance (one base + four variants). This is acceptable given V8's closure optimization for same-shape functions, and is the same pattern used by the top-level test() in harness.js.

Test plan

• Verify all variant functions exist on TestContext (including expectFailure)
• t.test.skip() prevents callback execution
• t.test.todo() without callback (marks as todo, skips)
• t.test.todo() with callback (runs callback, marks as todo)
• t.plan() counting works with subtest variants
• Nested subtests also expose all variants

Fixes: #50665

[nodejs-github-bot]

Review requested:

• @nodejs/test_runner

[Felipeness]

This is an alternative to #61251, addressing the same issue. Key differences:

• Includes expectFailure in the variants list (matching the full set from harness.js)
• Extracts a shared runSubtest() helper to avoid code duplication
• Additional test coverage: todo with callback, plan counting with variants, expectFailure existence check

[JakobJingleheimer]

Thanks. The DRY'ing up is an advantage over the original PR this came from; I think there is a better way to go about it though: #61251 (comment)

It's maybe not so nice that you copy+pasted an entire file from that other PR 😕 (I see you did cite the origin PR in a comment here)

[JakobJingleheimer]

Duplicate of #61251

[Felipeness]

Thanks for the review @JakobJingleheimer! I've addressed your feedback:

Changes in the latest commit:

1. Private method instead of constructor closure — Moved the subtest logic from a closure (runSubtest) inside the constructor to a private #runTest method on the class body. The constructor now only wires up this.test and attaches the keyword variants via ArrayPrototypeForEach, keeping it slim.

2. Rewrote the test file — Completely rewritten with original test names, structure, and assertions. Uses for..of loops to check variants instead of individual assertions. Added expectFailure coverage throughout (which test_runner: add skip, todo, only to subtest context #61251 is currently missing).

3. Sorted keyword list alphabetically (expectFailure, only, skip, todo) for consistency.

The approach follows the same pattern as harness.js: getCallerLocation() is called at the entry point (the arrow function the user invokes) and passed down to #runTest, preserving correct stack depth for source locations.

I'm aware this overlaps with #61251. Happy to close this in favor of that PR if preferred — the DRY improvements and expectFailure coverage here could also be contributed there as follow-up.