Use bic instead of and on arm for clang compatibility

[krovetz]

Hello Andrew,

I'm using your chacha implementation for a project and thought I'd report a very small issue.

Because the constants are too big for and, gcc is nice enough to assemble and sp, sp, #0xffffffe0 into the equivalent bic sp, sp, #31 and and sp, sp, #0xffffff80 into the equivalent bic sp, sp, #127. Clang however doesn't. Changing the source to use bic would make the code work on both compilers.

Thanks for the nice implementations!

Ted Krovetz