fixed bug for images with odd-numbered columns

[dmagree]

This fixes a problem when half-sampling images with an odd number of columns. Pointers top and bottom finish the for loop one pointer short if the out.cols does not evenly divide in.cols.

[cfo]

are you sure?
note that the stride is sometimes not equal to in.cols because of the internal memory layout.

[dmagree]

Hmm, I don't think it matters, does it? The problem with lines 111 and 112 are that top and bottom are not at the end of their respective rows. Stride will never advance them the correct amount in that case. The fix assumes that top and bottom are 1 byte away in the case of odd number of columns, which is consistent with how the rest of the section is written, correct?

[dmagree]

Ah, ok, but I should be checking in.cols instead of stride at 93. If you agree I'll fix and make another pull request.

[kvmanohar22]

@cfo This probably should have been merged quite a long ago. Please have a look at the valgrind report at #32