Fix 456 (unaligned memory for Hamming)

[erikbern]

This might fix it, @jlmelville wdyt?

[jlmelville]

Thanks for attempting this so quickly @erikbern. This passes the GCC-UBSAN check on Rhub, but it's hard to be certain of the fix because due to the stochastic nature of the original problem.

[erikbern]

cool. i'll try to address the failing travis build in the near future

[jayvdb]

note this causes segfaults in hamming_index_test.py::HammingIndexTest::test_many_vectors

[erikbern]

i completely forgot about this PR, will revisit shortly