Change Tests.prototype.all_done to require either tests or pending remotes

[andreubotella]

Tests.prototype.all_done used to treat the presence of at least one test as a necessary condition for all tests to be done. This would be false in cases where an exception is thrown inside the setup function, but the code path that runs in that case doesn't use the all_done() function.

However, if an exception is thrown inside the setup function in a worker test, tests.all_done() is called in that code path, which results in the completion message from the worker being ignored. This change fixes this by changing the condition to require the presence of either tests or remote contexts.

For service worker tests, however, that is not enough, since service worker registration is asynchronous and tests.all_done() will be called before the SW remote context is set up. This change lets the fetch_tests_from_worker function take a promise or an async function that returns the worker, and it guarantees that the test won't be considered complete until the promise is resolved. The HTML boilerplate for service worker tests is also changed in turn.

Closes #29777.

---

For historical context, this PR was initially a draft fix for #29777 that removed the condition that this.tests.length > 0. This turned out to break many tests (#29777 (comment)), and this PR was repurposed to solve this in a different way.

[andreubotella]

Closing because the amount of test breakage this change introduces (see #29777 (comment)) means it's probably best to try other options to solve #29777.

[andreubotella]

Reopening because replacing this.tests.length > 0 with (this.tests.length > 0 || this.pending_remotes.length > 0) might just do the trick.

[andreubotella]

PR rebased, squashed, renamed and marked as ready for review.

[andreubotella]

I've checked the diffs of this PR with the corresponding commit in master (7df27fb) for Chrome, and every difference seems to be due to a fluke that also affects other recent runs. The one exception is storage-access-api/hasStorageAccess.sub.window.html, but while debugging that I found bugs with the test itself (see #30320), and after fixing those bugs, the results before and after this change are identical (both for wpt serve and wpt run). So this PR by itself doesn't seem to be causing any unexpected failures or successes.

[jgraham]

No substantive concerns, but I think we need a docs update, and either an RFC or some agreement that this is a small enough change to skip an RFC.

[andreubotella]

I just realized that the change to make fetch_tests_from_worker take a promise isn't really needed anymore. I added it back when this PR removed the this.tests.length > 0 condition, since it made SW tests be marked as done before the registration promise resolved. But when I changed the condition to (this.tests.length > 0 || this.pending_remotes.length > 0), now a test file that only fetches tests from a SW wouldn't complete before the remote context is set, so there's no longer any reason to add a dummy pending remote while the promise is pending.

[andreubotella]

Checked the diffs against master again, and there are no regressions that can't be explained as flaky tests.

[jgraham]

Thanks for all the work on this. I know it's hard to get even a one line change to land :)