support building with libgmp as well as num-bigint

[arvidn]

the mpz_import and mpz_export functions are quite a bit slower than the num-bigint counterparts, so this slows down the benchmarks somewhat.

The rationale for this is:

1. use more main-stream code, with lower security risks
2. (possibly) use the same instance of libgmp as we already use in chiavdf and bls-signatures

[arvidn]

interestingly, there seems to be a rule preventing building the FFI layer on windows: https://github.com/Chia-Network/clvm_rs/runs/4189087870?check_suite_focus=true#step:7:104

That probably has to be addressed before we can try to get MPIR working.

[github-actions[bot]]

'This PR has been flagged as stale due to no activity for over 60 days. It will not be automatically closed, but it has been given a stale-pr label and should be manually reviewed.'