Inline fixed-width PUSH immediates

[Giulio2002]

Summary

• generate fixed-width inline cases for PUSH5-PUSH19 and PUSH21-PUSH31
• avoid the generic uint256.Int.SetBytes path for these static immediate widths
• keep existing specialized PUSH0-PUSH4, PUSH20, and PUSH32 cases

Benchmark

Focused BenchmarkOpcode PUSH cases, compared against the pre-change branch run:

• Opcode/PUSH5_POP: 9.908ms -> 6.657ms, -32.82%
• Opcode/PUSH16_POP: 9.806ms -> 6.659ms, -32.09%
• Opcode/PUSH31_POP: 10.193ms -> 8.718ms, -14.47%
• Opcode/PUSH1_POP: no meaningful change

Verification

• go generate ./vm
• git diff --check
• go test ./... -count=1
• GEVM_TEST_RUNNER=opcode go test ./... -count=1
• make test
• go test ./tests/bench -run '^$' -bench '^BenchmarkOpcode/(PUSH1_POP|PUSH5_POP|PUSH16_POP|PUSH31_POP)$' -benchmem -count=8

[Giulio2002]

Closing this version; fixed-width PUSH was reverted. Replacing it with a smaller POP codegen optimization PR.