Reduce number of dependencies of llrt_crypto

[Sytten]

The module has too many dependencies, it really pollutes any project trying to import it since some have very deep web of dependencies. For example we pull in two different bigint implementations (num-bigint-dig and crypto-bigint)
We have a lot of duplicate with ring anyway and we should prefer using it. Curve crypto is still not in ring, but planned (briansmith/ring#1771)

[richarddavison]

I shared your concern @Sytten when starting the crypto refactor derived from @nabetti1720 fantastic ground work. Problem with ring is that it only supports a fraction of the crypto APIs/algos we need. Even if ring has support for an algorithm, it only supports decryption/verification and not encryption/signature etc.
That being said, we should provide different web crypto implementations:

• One with pure Rust implementations: using the current crates
• Another one with system: OpenSSL/LibreSSL/BoringSSL

[Sytten]

We might want to look at graviola (https://crates.io/crates/graviola) made by a maintainer of rustls.

[richarddavison]

That would be great. Unfortunately we can't get rid of everything as graviolia/ring/aws-lc only supports some of the operations we need in subtle crypto. I don't think there is plans to support full API either:
briansmith/ring#691
To my recollection aws-lc-rs also only exposes the same APIs as ring and does not have bindings for the full lib crypto APIs. Otherwise that would be an option.