Task cancellation within a test is handled inconsistently

[grynspan]

Today, the task cancellation APIs behave inconsistently in Swift Testing:

• UnsafeCurrentTask.cancel() within a test body cancels the current test case but has no other observable effects
• UnsafeCurrentTask.cancel() during trait evaluation will cancel the entire test run (!)
• try Task.checkCancellation() after task cancellation will record an error even though the test's task has been cancelled

Task.cancel() and UnsafeCurrentTask.cancel() don't throw, so they don't implicitly provide for early termination, but they can be paired with throw CancellationError() or even just return.

The only reasonable behaviour for task cancellation in a test target is that, if you cancel the task associated with a test, it cancels the test.

[grynspan]

Tracked internally as rdar://159150449.