Add faster reduction for secp256k1 pcurves

[randombit]

This takes advantage of the modulus being near a power of 2.

Rough idea of improvements. Seems to vary a lot between compilers and CPUs:

• Sandybridge/GCC: ECDH and ECDSA all about 9% faster
• Skylake/GCC: ECDH 19% faster, ECDSA sign 13% faster, ECDSA verify 17% faster

#4027

[coveralls]

coverage: 91.771% (+0.005%) from 91.766%
when pulling 4cd489f on jack/faster-k1-redc
into 856174e on master.

[coveralls]

coverage: 91.77% (+0.004%) from 91.766%
when pulling 8839e0a on jack/faster-k1-redc
into 856174e on master.

[coveralls]

coverage: 91.78% (+0.01%) from 91.766%
when pulling 5815410 on jack/faster-k1-redc
into 856174e on master.

[coveralls]

coverage: 91.776% (-0.006%) from 91.782%
when pulling c33c00f on jack/faster-k1-redc
into 4c3e9dc on master.

[coveralls]

coverage: 91.779% (-0.003%) from 91.782%
when pulling efdbdc8 on jack/faster-k1-redc
into 4c3e9dc on master.

[coveralls]

coverage: 91.727% (+0.001%) from 91.726%
when pulling ac87424 on jack/faster-k1-redc
into 7fad1d2 on master.