Add support for float16 matmul

[LaurentMazare]

As mentioned in #41 it would be nice to have support for half precision matmul (especially when using mkl).
This PR adds it and goes together with blas-lapack-rs/blas-sys#21 on the sys side. The blas-sys version has been bumped to reflect the new requirement.

Let me know if there is anything that could be done to improve this PR or the blas-sys one, or if you see any different way to get this working.
Thanks,

Note that this has been tested in a very simple way via the following:

extern crate intel_mkl_src;

use half::f16;

fn main() {
    println!("Hello, world!");
    let alpha = f16::ONE;
    let beta = f16::ZERO;
    let a = vec![f16::from_f64(0.5)];
    let b = vec![f16::from_f64(0.5)];
    let mut c = vec![f16::from_f64(0.5)];
    unsafe {
        blas::hgemm(
            b'N',
            b'N',
            1,
            1,
            1,
            alpha,
            a.as_slice(),
            1,
            b.as_slice(),
            1,
            beta,
            c.as_mut_slice(),
            42,
        )
    };
    println!("{c:?}");
}