feat: optimized secp256k1 ECDSA verification
3 algorithmic optimizations over zig stdlib, no assembly:
1. endomorphism via 1 field multiply (not ~65 doublings)
2. single 4-way Shamir loop (128 doublings, not 256)
3. projective-space comparison (no field inversion)
3.3x faster than stdlib on 3072-entry atproto corpus.
drop-in API compatible with std.crypto.sign.ecdsa.EcdsaSecp256k1Sha256.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>