i386 SHA1Transform.

This version has the same speed (maybe marginally better... 2~5%), is more compact (3.65 Kb of code down from 4.25, 1254 instructions down from >1440), and most importantly, it might resolve #38260 (closed). If it indeed fixes everything, please re-enable this thing for all i386 platforms yourselves. If it does not, you haven’t lost anything.