Switch inlining to subroutines in non-alt ARM P-521 point doubling

The previous version with the field operations inlined led to both a
very long-running proof and rather extravagant code size. The big two
field operations (modular multiplication and squaring) now call local
subroutines instead, which solves both problems and seems to have
minimal impact on performance.

s2n-bignum original commit: awslabs/s2n-bignum@8c27977

s2n-bignum original commit: awslabs/s2n-bignum@61e4747