k256: add impl of ReduceNonZero<U256> for Scalar

[fjarri]

• add impl of ReduceNonZero<U256> for Scalar
• fix a bug in WideScalar::reduce_nonzero() (a zero value returned when reducing MODULUS-1)

Very embarrassing for me, but at least chances of someone hitting it were astronomically small, and if they did, it most probably resulted in a panic during inversion or something of the sort.

[tarcieri]

there may be a small performance penalty for calculating order_minus_one in the method itself

It's const fn so I'd hope not, although binding it to a function-scoped constant would be a way to make sure.

should I add tests?

That'd be nice 😉

[fjarri]

It's const fn so I'd hope not, although binding it to a function-scoped constant would be a way to make sure.

Fixed.

Also it's a good thing you asked me to add tests...

[tarcieri]

Thank you!

[fjarri]

Btw, are you planning on making a release of k256 any time soon? My other project would benefit from having this available.

[tarcieri]

Sure, I can cut a release

[tarcieri]

In case you didn't see, I cut k256 v0.10.1 yesterday.

Another elliptic-curve crate release might take a bit longer as I'd like it to include hash-to-curve work, which is currently half-finished.

[fjarri]

Thanks, appreciate it!