IECoreScene tests : Tighten up a couple test timings

[danieldresser-ie]

I'm not sure what the correct solution here is long term. John suggested simply removing the wait entirely, and setting the canceller as cancelled before launching the thread we are cancelling. That doesn't really test the right thing though ... we want to make sure that something is still listening for cancellation once the code has had a chance to get into an inner loop.

Without confidence in an ideal solution, I've just made the test timing much tighter, which is probably enough to get the tests passing.

In the case of Triangulate at least, I suspect that what went wrong may be related to actual optimizations that reduced the runtime of the triangulate enough that the wait was too long.

I'm hoping that these values are good enough to stop seeing failures currently, though I'm open to other proposals for the long term.

[johnhaddon]

Thanks Daniel. Rebased to retarget to RB-10.5, and merging.