Optimization in alignment.

[andrewdavidsmith]

This should fix a bug affecting certain reads on macOS Ventura, intel and gcc-13 with O3 optimization.

I added directives for compiler to do less aggressive vector optimization around the tight loops that iterate in parallel while doing alignment in a band. Specifically, the from_diag, from_above and from_left, both those with and without traceback. I was able to alter only a subset of those 6 functions and make certain test cases work, but the only way the entire regression test file in dnmtools would pass is with all 6 functions allowing at not more than the very-cheap option for dynamic vectorization. It seems like a second directive can turn the optimization back on for functions below in the file.