Save 1 instruction in libdivide_u*_branchfree_do()

[kimwalisch]

As discussed in https://github.com/ridiculousfish/libdivide/issues/12 I have made a modification to libdivide_u*_branchfree_do() so that it uses 1 instruction less.

There are 2 possible ways to implement this:

• Make minor modifications in libdivide_u*_branchfree_gen(), libdivide_u*_branchfree_recover() and libdivide_u*_branchfree_do().
• Implement a separate libdivide_internal_u*_branchfree_gen (probably some other functions need to be duplicated as well).

I chose the first solution because it requires less code and is much easier to implement. Though I have to agree that the second solution would be cleaner at the expense of duplicating some code.

Please let me know what you think about these code changes.

[ridiculousfish]

Thanks!