Add support for bw6-767

[seunlanlege]

Builds on the work previously started here #47

Summary

• Generic curve infrastructure: Refactor the entire codebase (keyset, prover, verifier, setup, BLS, transcript, endomorphism, PIOP) to be generic over curve types and PCS, replacing the hardcoded BLS12-377/BW6-761 pairing. Introduces an instances module with concrete configurations for both BLS12-377/BW6-761 and BLS12-381/BW6-767.
• Smooth-domain FFT for low two-adicity fields: Implement SmoothSubgroupDomain for fields like BW6-767 Fr (TWO_ADICITY=1), using Good-Thomas + Rader's and Cooley-Tukey + Rader's mixed-radix FFT strategies. Domain size 1034 (2×11×47) for BW6-767.
• Generic evaluation domain: Thread D: EvaluationDomain<F> through all core structs so both Radix2 (BW6-761) and SmoothSubgroup (BW6-767) domains work transparently.
• Fix non-power-of-2 domain assumptions: Fork fflonk to remove TWO_ADICITY check in KZG SRS generation, ensure 2x/4x domains are proper multiples of base domain size, and remove power-of-2 assumptions in PIOP (rotate_left, vanishing poly eval, last-point indexing).
• Generalize bitmask packing: compute_block_size() finds the largest divisor of domain size that fits in the field bit capacity, replacing the hardcoded 256-bit block size.
• Optimize Rader's FFT: Performance improvements to the Rader's algorithm implementation.
• Migrate to arkworks 0.5 and fflonk 0.0.3

Results

BW6-767 APK proof: 773 bytes, keyset size 1033, domain size 1034.

Testing

• cargo test passes for existing BW6-761 tests
• bw6_767_apk integration test passes
• Bitmask packing test verifies BW6-767 block=94, 11 chunks
• Endomorphism constants verified via compute_endo test