Fix toplevel window resize pingpong

[tarek-y-ismail]

closes #3573.

[AlanGriffiths]

I think this approach is wrong. In essence, it is guessing whether the client has reacted to a configure event by looking at the current size. That will never work for all cases.

As we have sent a configure, we should be tracking the ack_configure (similar to the layer-shell logic you just fixed)

[AlanGriffiths]

(And, yes making the test more complex is tempting. I fell for it when I looked at #3040 a year ago)

[tarek-y-ismail]

So I think the "loop" was intentional, but the guard clause doesn't work because requested_size isn't cleared after the client size is set, thus checking against the OLD value if requested_size on the next iteration (which continuously flip-flops) instead of checking against the window size. Instead, since requested_size is used as a sort of double buffering, it should be cleared at some point (when window_size is updated for example`)

[AlanGriffiths]

As previously discussed, I don't think this is right, but it does improve things and doesn't seem to cause additional problems. (In particular, canonical/ubuntu-frame-osk#58 remains fixed.)

(I have a suspicion that there may be more conflation of window_size and content_size around that needs resolving for SSD, but that's out of scope for this PR)