🧠 What's inside UltrafastSecp256k1:
• BMI2/ADX asm → 3-5× speedup
• CUDA batch: 4.63M keys/sec
• Montgomery batch inverse
• GLV endomorphism
• Zero heap allocs in hot paths
• Constant-time ct:: namespace
• All primitives from scratch
No boost. No openssl. No libsecp256k1. Just C++20.

