Wait for the pending candidate before updating the pointer

These tests mutate the media element while the resource selection algorithm is
parked on a candidate, then check which sources get processed. They did the
mutation from a script that runs during parsing. Per spec that is late enough:
"await a stable state" queues a microtask, and the parser performs a microtask
checkpoint at </script>, so the pointer is (b | c) with b pending by then.

No engine resumes there. Firefox Nightly, Chrome Canary and Safari TP all run
the synchronous section in a task, so the mutation landed before the algorithm
had started and the tests measured tree order rather than the pointer update
rules. Two of them failed for that reason alone
(pointer-insert-source and pointer-remove-source, on x1); the other four
happened to expect the same result under either timing, so they passed without
testing anything about the pointer.

Wait for video.currentSrc to become source b's URL instead. currentSrc is set in
the "process candidate" step, after the pointer has advanced and before the
resource fetch algorithm runs, so it directly observes the state these tests
need, under either timing. Polling is used rather than an event because no event
works: loadstart is queued before the mode-specific steps, and engines disagree
on whether a source's error task is dispatched before or after the pointer
advances. Each test also asserts that b is still pending at that point, so a
wrong sync point fails loudly instead of silently measuring something else.

pointer-remove-text.html additionally iterated a live childNodes while removing
from it, which skipped entries and only removed every text node by coincidence
of the alternating markup. Snapshot the list and assert the removal happened.

pointer-remove-source-after.html did its four removals in four separate
parse-time scripts, which under the new sync point collapse into one step.

Firefox Nightly now passes all six, including the two that failed before.
Chrome Canary and Safari TP still fail all six on "error events on c", which is
unrelated: both fail the unmodified pointer-control.html the same way, because a
source with no src that follows a failed src-bearing source never fires error.