Optimize MLOAD memory fast path

[Giulio2002]

Summary

• avoid helper calls in the inlined MLOAD fast path by reading directly from the current memory buffer
• skip recopying bytecode in ResetWithHash when the pooled bytecode already has the same hash and length
• regenerate the VM dispatch table

Benchmarks

• BenchmarkOpcode/MLOAD: 9.766ms -> 8.926ms (-8.60%)
• BenchmarkDataMovement/KeccakSizes geomean: 14.00ms -> 13.49ms (-3.64%)

Tests

• go test ./... -count=1
• git diff --check