Refactored MLDSAParams to remove the hideous turbofish.

[ounsworth]

Solution: #105 introduces a really nice pattern for encapsulating AES params, and I want to also apply this to the ML-DSA and ML-KEM crates.
The pattern is to hide all the params in a pub trait, with a private trait that seals it.

The entrypoint to start the review is: the new trait MLDSAParams in params.rs.

This allows for greatly simplifying the previous turbofish construction, which was getting quite out of hand, brittle, and easy to make a mistake.

It used to be:

pub type MLDSA44 = MLDSA<
    MLDSA44_PK_LEN,
    MLDSA44_SK_LEN,
    MLDSA44_SIG_LEN,
    MLDSA44PublicKey,
    MLDSA44PrivateKey,
    MLDSA44_TAU,
    MLDSA44_LAMBDA,
    MLDSA44_GAMMA1,
    MLDSA44_GAMMA2,
    MLDSA44_k,
    MLDSA44_l,
    MLDSA44_ETA,
    MLDSA44_BETA,
    MLDSA44_OMEGA,
    MLDSA44_C_TILDE,
    MLDSA44_POLY_Z_PACKED_LEN,
    MLDSA44_POLY_W1_PACKED_LEN,
    MLDSA44_LAMBDA_over_4,
    MLDSA44_GAMMA1_MINUS_BETA,
    MLDSA44_GAMMA2_MINUS_BETA,
    MLDSA44_GAMMA1_MASK_LEN,
>;

And it's now:

pub type MLDSA44 = MLDSA<
    MLDSA44Params,
    MLDSA44PublicKey,
    MLDSA44PrivateKey,
    MLDSA44_PK_LEN,
    MLDSA44_SK_LEN,
    MLDSA44_SIG_LEN,
>;

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

[dghgit]

Code passes sanity check, nice reduction, some reduction in public API as well.

Fable review was okay, I've sent you the report, there was one possible regression and a couple minor things picked up. I think they all predate this PR though.

[dghgit]

Good to see. I wonder if there's some one we can tell our electronic friends to keep a watch out for patterns like this as well, it's the kind of thing that's very easy to do, even in a post-LLM world.

[dghgit]

Merged on release/0.1.3alpha