Migrate mod_exp to BoringSSL, removing GMP requirement

[bhazzard]

As part of the effort to remove system provided "mathy" dependencies that could unexpectedly change out from under us and affect protocol consensus, remove the sole remaining usage of libgmp: the modexp host function. This likely can be refactored to use boringssl fairly easily, and that then opens up the possibility for #1255 too since we can maintain a fork with the needed hooks.

We need to be mindful to measure performance difference between the two; and pay special attention that boringssl has different optimization paths depending on the input parameters (this may mean we ought to create specialized benchmarks for such inputs).

We must add unit tests for all pertinent modexp test cases that are part of boringssl and ensure they are equivalent to the existing consensus rules. The output of boringssl/openssl's modexp has changed over time to rectify implementation defects and while the expectation is gmp's modexp is the same as boringssl at the current time, these unit tests exercise some fantastic corner cases. These unit tests should be added in a discrete PR, or at least pushed individually, for easy verification that the results are as expected on gmp.

[bhazzard]

Unit tests could potentially be a separate PR.

[BenjaminGormanPMP]

Issue moved from iteration to backlog. Per @spoonincode - this issue is dependent on #644.

[greg7mdp]

Wondering if Boost multiprecision could be used?

[spoonincode]

Wondering if Boost multiprecision could be used?

https://www.boost.org/doc/libs/1_83_0/libs/multiprecision/doc/html/boost_multiprecision/perf/integer_performance.html
table 1.71 makes it look like the non-GMP impls Boost MP supports are slower than GMP too

[greg7mdp]

Ha, that's right. Not quite as bad as your Boring results though. Not sure how critical this is.