Fix false positives caused by cover conditions that are never true

[bgerstle]

What's in this pull request?

Testable.cover would fail to add coverage labels if the condition was always false, leading to false "sufficient coverage" test results. Take the added regression test in PropertySpec as an example:

// CHECK-NEXT: +++ OK, failed as expected.
// CHECK-NEXT: *** Insufficient coverage after 100 tests
// CHECK-NEXT: (only {{[0-9]+}}% never, not {{[0-9]+}}%)
property("Cover reports failures for conditions that never happen") <- forAll { (s : Set<Int>) in
    let somethingHappened = false
    return (s.count == [Int](s).count).cover(somethingHappened, percentage: 1, label: "something happened")
}.expectFailure

The expected failure never occurs because somethingHappened is always false and, as a result, the `"something happened" label was never added.

This PR fixes the cover logic so that the labels are always upserted, but the stamp is only added if the condition is true. With this change, the above test passes because cover fails the test due to insufficient coverage ("something happened" only 0%, not 1%).

Why merge this pull request?

Fixes the expected behavior of cover in an edge case.

What's worth discussing about this pull request?

Whether the changes to cover could cause unintended side effects, as it's reused in multiple other cases (classify, label, etc.). I'd be willing to add more tests if people think of other scenarios worth checking.

What downsides are there to merging this pull request?

None that I can think of.

[CodaFi]

LGTM

⛵️

[CodaFi]

⛵️