Let RngCore: TryRngCore<Error = Infallible>

[dhardy]

• Added a CHANGELOG.md entry

Summary

Revise the core traits according to @newpavlov's suggestion and similar to #41 but keeping the infallible methods.

No trait renames are done in this PR. Also not sure if we need any.

Details

• TryRngCore is implemented for any R: DerefMut where Target: TryRngCore.
• RngCore: TryRngCore<Error = Infallible> with automatic implementation over TryRngCore
• UnwrapMut is removed
• CryptoRng = RngCore + TryCryptoRng or as close as possible without sum traits; it is kept since I believe some people need dyn CryptoRng
• utils::{next_u64_via_u32, next_word_via_fill} now use closures instead of R: RngCore

Questions

Should utils fns support potentially-fallible uses (try_next_u64_via_u32 takes FnMut() -> Result<u32; E> for generic E)? I think unnecessary.

[dhardy]

Updated. Some things need further discussion (see above).

[tarcieri]

We noticed something a little weird here:

• Bump rand_core to v0.10.0-rc-4 RustCrypto/traits#2195
• elliptic-curve: fixup rand_core API use RustCrypto/traits#2197

So:

pub trait CryptoRng: TryCryptoRng<Error = Infallible> {}

impl<R: TryRngCore<Error = Infallible>> RngCore for R { ...

I would expect that means any type which impls CryptoRng also impls RngCore, but in RustCrypto/traits#2197 we had to add explicit RngCore bounds (or reborrow as in RustCrypto/traits#2195)

[newpavlov]

I guess we should revert my suggestion above and add a test for this.

[tarcieri]

I don't get it... I tried to make a contrived version in the playground and it... works?

https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=be7a3fdc3b7bc93fb22c1529d99d4f03