k256: normalize before calling is_odd() in sng0()

[fjarri]

An is_odd() call I forgot to address in #530

While this fixes the problem at hand, there are a couple of considerations:

• To prevent future problems like this, we do need either k256: type safety for non-normalized field elements #531, or just auto-normalizing in FieldElement::is_odd() (but not in FieldElementImpl!). Originally I didn't do that because I wanted to leave a possibility for a slightly more efficient code (where the caller would know whether the value has been previously normalized or not), but perhaps it is not worth it, and it's better to prevent errors instead.
• This was not caught in tests in k256: Do not normalize the argument in FieldElementImpl::is_odd() #530 because in CI we run tests with --release, and apparently this bug does not affect hash2curve test, which is the only one touching sgn0(). @tarcieri , why was that decision made originally?

[tarcieri]

in CI we run tests with --release, and apparently this bug does not affect hash2curve test, which is the only one touching sgn0(). @tarcieri , why was that decision made originally?

Are you asking why CI runs the tests with --release?

We've had bugs in the past that only affected the release profile. An easy way to screw up is to perform some sort of effectful operation inside of a debug_assert! which is completely elided in a release build.

Release builds are what people actually use, so we test against them as they're more important.

We can potentially test both debug and release builds if you think that'd be helpful.

[fjarri]

We can potentially test both debug and release builds if you think that'd be helpful.

I'm actually not sure now how it happened, because FieldElementImpl safety harness should have still been active when CI ran in #530 - but I see hash2curve test passing there, but if I run cargo test --all-features locally it fails. Yesterday I didn't notice it because I only ran cargo test with default features locally.

[fjarri]

I hope it's not too much to ask to publish k256 0.10.4 with this? My library is using hash2curve, that's how I noticed this - in my own tests.

[tarcieri]

v0.10.4 released: #534