Resample Optimization

[danieldresser-ie]

Minor optimization of Resample, fixing cases where we were traversing more pixels than necessary, and while I'm at it, pulling the calls to floor/ceil out of the inner loop.

Performance results are:

- testPerfInseparableAwkwardSize (GafferImageTest.ResampleTest.ResampleTest) : was 2.00s now 0.74s (63% reduction)
- testPerfInseparableDisk (GafferImageTest.ResampleTest.ResampleTest) : was 6.31s now 5.92s (6% reduction)
- testPerfInseparableLanczos (GafferImageTest.ResampleTest.ResampleTest) : was 15.16s now 14.44s (5% reduction)
- testPerfSmallFilter (GafferImageTest.ResampleTest.ResampleTest) : was 0.43s now 0.39s (8% reduction)
- testPerfVerySmallFilter (GafferImageTest.ResampleTest.ResampleTest) : was 0.39s now 0.36s (10% reduction)

Also, now that a box filter just samples the values it uses, rather than also taking adjacent pixels and multiplying them by zero, it no longer converts inf to NaN, so we can remove the workaround from MergeTest.

[johnhaddon]

LGTM. I've pushed a rebase just to fix the conflicts in Changes.md - everything else is the same. Merging.