Implement OpenMP-based parallelism for C lib

[ibmibmibm]

Mimics all TBB changes (#445).

Related issues: #229 #184

[oconnor663]

I don't love the prospect of maintaining two separate implementations of this, but I'm curious to hear your thoughts about how many callers might prefer/demand to use OpenMP instead of TBB.

Separately, there was some discussion in the TBB PR about how OpenMP might not perform well with the nested/dynamic task spawning strategy that we're using here. @sylvanshade commented that:

...OpenMP is generally not suitable for nested parallelism, which is exactly how the problem is being solved here with Rayon and the oneTBB implementation. Most of the performance recommendations I see regarding OpenMP suggest to avoid nested parallelism at all costs because it's known to be inefficient.

@sneves made some performance measurements that seemed to corroborate @sylvanshade's comment. And when I run the benchmarks you added to blake3_c_rust_bindings (thank you for being so thorough!), it looks like OpenMP is having some trouble:

$ cd c/blake3_c_rust_bindings && cargo +nightly bench --features=tbb,openmp 1024_kib
...
test bench_incremental_1024_kib        ... bench:     117,774.41 ns/iter (+/- 1,743.86) = 8903 MB/s
test bench_openmp_1024_kib             ... bench:     946,225.39 ns/iter (+/- 507,769.54) = 1108 MB/s
test bench_tbb_1024_kib                ... bench:      47,784.35 ns/iter (+/- 6,406.28) = 21944 MB/s

Have you seen anything different in your measurements? Are there other cases we should be testing?