Improve secp256k1 field multiplication speeds

[tuxxy]

Presently, the k256 crate implements montgomery mulmod for multiplication/reduction. @hdevalence brought up a point in #19 that there may be a method that leverages the special form of the secp256k1 modulus to improve multiplication performance.

Performance of a scalar multiplication for the k256 crate:

test tests::k256_scalar_mul    ... bench:     197,252 ns/iter (+/- 2,838)

It would be nice to see a significant performance increase for this.

[tarcieri]

@tuxxy would you consider this addressed by #59?

[tuxxy]

Not quite. I'd like to write some benchmarks for myself and talk with @fjarri a bit before it gets closed.

[fjarri]

What multiplication are we talking about here? Scalar-scalar, field-field, or point-scalar? The latter can certainly be improved, I'm planning to look into it.

Where's this benchmark you're quoting, by the way?

[tuxxy]

I've unfortunately lost the benchmark I originally wrote, but it measured point-scalar. It would be nice to see two benchmarks, 1) comparing point-scalar operations to libsecp256k1 and 2) measuring scalar-scalar operations (maybe against a variety of implementations?).

[tarcieri]

I think we can go ahead and close this out. Further optimizations would certainly be appreciated, but I think we're ok for now.