Cancel existing operation in a watcher before starting a new one

[designatednerd]

Thanks to @RolandasRazma for the suggestion in #1011 - this removes the hack needed to make things pass added in #1010.

There's still some pretty significant thread safety issues going on here as pointed out by @cysp, but I think there's a BIG rabbit hole there, and I'd at least like to remove the dirty hack ASAP.

[cysp]

Cancelling any previous fetch definitely seems like something that should be done.

I guess no part of the test suite asserts the value of that state property?

[designatednerd]

@cysp It doesn't appear that we have tests specifically for either AsynchronousOperation or for FetchQueryOperation, which seems to be where the main issue is. We also don't have TSAN turned on for tests in general because there's a bunch of tangled stuff causing races we haven't (so far) been able to detangle. Again, any help we can get on that would be much appreciated.