Performance optimizations: LLVM function ptr caching, CGO handle replacement, and native memory opcode

[dajuguan]

• Cache function pointers → ~1000 ns performance improvement for WETH contract
• Replace cgo handle with Go function map → ~100 ns improvement for WETH contract
• Implement Memory opcodes in C and LLVM → ~500 ns improvement for WETH contract

Detailed benchmark: #82