*Significant* MD5 speedups
This change requires a new Jambase, so, a new Jam, to take full
advantage.
Jambase: Now we handle local variables OPTIM, SUBDIRC++FLAGS, and
SUBDIRCCFLAGS on targets correctly. This was a change contemplated
ages ago on the jamming@ mailing list but never committed.
md5.cc: We no longer do byte-swabbing on temporary buffers,
we use a union to treat buffers as char * or uint32 *. We have
re-ordered the MD5STEP macros to be faster slightly based on
Wei Dai's "Public Domain" crypto++ 5.6.1 routines.
md5.h: We now align our buffers (per Wittenberg's recommendation) for
those machines both where it's faster to do so. Windows will be next?