fix segfault related to #128

[rob-p]

This needs some more (independent) validation, but this should fix the (long standing) segfault that previously was solved with the repeated km_init hack.

Essentially, the previous code was handing out multiple references (including a mutable one) to the underlying thread local buffer at the same time. This voilates the rust shared xor mutable rules and, recalling that unsafe doesn't disable the rules and that we are still responsible for upholding them, I believe that opened us up to undefined behavior ultimately resulting in the segfault.

This PR applies an easy fix. It simply moves the acquisition of the mutable pointer to the km object until after the call to mm_map, so that we only acquire the mutable reference after the lifetime of the mm_map call is finished with it. In my testing, this seems to resolve the segfaults that were previously occurring.