feat: use openvm-kzg for KZG point evaluation precompile

[github-actions]

Adds support for KZG precompile.
Use of intrinsics for BLS12-381 is:

• feature gated by "kzg-intrinsics" on guest
• enabled by cmd line arg --kzg-intrinsics on host

Default behavior: does not turn on "kzg-intrinsics".

On block 21,000,000 where no kzg point eval precompile is called:

• without kzg-intrinsics: https://github.com/axiom-crypto/openvm-reth-benchmark/blob/gh-pages/benchmarks-dispatch/refs/heads/update-openvm-20250218223742/reth-d8a4152cf8ee045cab3c8363b0d8c5c0683ae2c9-94418a3a55787953df9cc1e8a7306183acd7540f4ef1f2c684279233395c672d.md
  • total proof time: 633.5s (nearly no change from main)
  • par proof time: 185.5s (main is 182s with variance)
• with kzg-intrinsics: https://github.com/axiom-crypto/openvm-reth-benchmark/blob/gh-pages/benchmarks-dispatch/refs/heads/update-openvm-20250218223742/reth-ed650ac019745936a133473b914cea4296db7fce-94418a3a55787953df9cc1e8a7306183acd7540f4ef1f2c684279233395c672d.md
  • total proof time: 710.4s
  • par proof time: 189s

On block 21882667 where kzg point eval is called once:

• without kzg-intrinsics: https://github.com/axiom-crypto/openvm-reth-benchmark/blob/gh-pages/benchmarks-dispatch/refs/heads/update-openvm-20250218223742/reth-d8a4152cf8ee045cab3c8363b0d8c5c0683ae2c9-d6d2fa728ab3dcf7eeaf20f4d68423a50aeafce627b00d0db4cbaef4104fa85c.md
  • total proof time: 861.41s
  • par proof time: 182.5s
• with kzg-intrinsics: https://github.com/axiom-crypto/openvm-reth-benchmark/blob/gh-pages/benchmarks-dispatch/refs/heads/update-openvm-20250218223742/reth-ed650ac019745936a133473b914cea4296db7fce-d6d2fa728ab3dcf7eeaf20f4d68423a50aeafce627b00d0db4cbaef4104fa85c.md
  • total proof time: 539.9s
  • par proof time: 220s (note this is higher than without kzg-intrinsics!)

The problem with kzg intrinsics is that the leaf verifier's program code size goes up from 4M to 16M rows because BLS chips have a large number of constraints.

Closes INT-3333