tool.scheduler.unit_tests seems to be broken on ARM/AArch32 since bd7f3eb (@derekbruening, May 2023).

The problem seems to be that memref_is_nop_instr calls decode in a way that always decodes as A32 so a T32 NOP instruction is not recognised. So I can make it pass by inserting if (*(uint *)record.instr.encoding == 0xe320f000) return true; into that function.

Would it make sense to rewrite memref_is_nop_instr so that instead of using the decoder it just recognises the encodings, like how next_record works? Or is there a better way?