Add PrimeFieldBits support to Scalar

[kayabaNerve]

The char_le_bits function is a hack I don't legitimately recommend. I just felt it'd be better to make a PR, which works, than to solely file an issue requesting it as a feature.

[tarcieri]

I think this should work but haven't tested it

Suggested change

	let mut bytes = (Scalar::ZERO - Scalar::ONE).to_repr();
	bytes[0] += 1;
	debug_assert_eq!(DScalar::from_bytes_mod_order(bytes), DScalar::ZERO);
	bytes.into()
	crate::constants::BASEPOINT_ORDER.into()

[kayabaNerve]

I'll test it now, thanks.

[kayabaNerve]

That's a Scalar, not a byte array. Doesn't that break safety to have an unreduced Scalar in system? Especially now that Scalar addition assumes only unreduced Scalars exist? I'd guess it doesn't in practice, as it won't have any shouldn't never be set bits set, yet still sounds like something which should be a [u8; 32].

[kayabaNerve]

to_bytes().into() works, will push that.

[tarcieri]

Doesn't that break safety to have an unreduced Scalar in system?

Yeah, seems a bit odd to have it in the public API if it's unreduced /cc @rozbb

[pinkforest]

BASEPOINT_ORDER will be marked deprecated from 4.1.1 - this will be BASEPOINT_ORDER_PRIVATE as pub(crate)

• Deprecate BASEPOINT_ORDER from pub API consts #581

[kayabaNerve]

Sorry for this PR being a mess. I'm unsure why the no-std CIs are failing.

[tarcieri]

It's possible that bitvec isn't working on no_std for whatever reason? Though I'm not sure why there isn't a proper compiler error.

Regardless it would probably be good to split this out into a bits feature that enables ff/bits so that bitvec doesn't become a mandatory dependency for users of the group feature.

[kayabaNerve]

I use ff, with bits, on libraries with no-std targets so that shouldn't be it... (though I may technically solely test them on alloc, not no-std).

I'll make bits a feature. group-bits for clarity? It's the bits of dalek's group functionality (despite technically affecting ff).

[pinkforest]

It's because we don't do additional cargo hack of the features via std build and we only do it via no_std that detects any errors via features - e.g. red herring that just appears via no_std run which has the feature tested but not in std run we don't do additional duplicate run :)

[rozbb]

This looks good to me, thanks! Could you add an item in the changelog for this feature?

[kayabaNerve]

Thanks for moving this forward before I had the chance, rozbb :) And thanks pinkforest for the actual edit needed.