[Cryptech Core] fmc updates

[Paul Selkirk]

I finally merged fmc_clk_60mhz to master, and moved the FMC
byte-swapping from software to hardware.

The first part (fmc_clk_60mhz) speeds up rsa_1024 signing by about 12%
(about 7.5ms).

The second part speeds it up by a further 0.3% (about 200us), and allows
us to use memcpy() rather than a word-copy loop. However, it turns out
that calling memcpy() wipes out that saving. So I replaced memcpy() with
an inline loop-unrolled version. Rob commented "Pretty sure that gcc
automatically inlines memcpy() unless you ask it not to do that." So
maybe it's not the function call overhead, maybe it's the alignment
checks, who knows.

Releng has been updated, but I don't know if tarballs are generated
automatically or manually.

				paul