Fix UI8 assign on Linux

[yuxiaomao]

Fixes #142 , fixes #658 for me without breaking our game on Linux (Ubuntu 22.04).

I'm not very sure what I'm doing, so here is some description:
I found that Linux & Windows have different set of CPU registers (if I use the same cpu set, I have segfault)

hashlink/src/jit.c

Lines 227 to 239 in 8bc5f8f

	# ifdef HL_WIN_CALL
	# define CALL_NREGS 4
	# define RCPU_SCRATCH_COUNT 7
	# define RFPU_SCRATCH_COUNT 6
	static const int RCPU_SCRATCH_REGS[] = { Eax, Ecx, Edx, R8, R9, R10, R11 };
	static const CpuReg CALL_REGS[] = { Ecx, Edx, R8, R9 };
	# else
	# define CALL_NREGS 6 // TODO : XMM6+XMM7 are FPU reg parameters
	# define RCPU_SCRATCH_COUNT 9
	# define RFPU_SCRATCH_COUNT 16
	static const int RCPU_SCRATCH_REGS[] = { Eax, Ecx, Edx, Esi, Edi, R8, R9, R10, R11 };
	static const CpuReg CALL_REGS[] = { Edi, Esi, Edx, Ecx, R8, R9 };
	# endif

Esi, Edi are used in Linux, however they are not 8 bit register (remove the id check on RCPU fixes the issues, but will make our game to exit "normally" without actually run and I don't understand why)

hashlink/src/jit.c

Lines 542 to 544 in 8bc5f8f

	static bool is_reg8( preg *a ) {
	return a->kind == RSTACK || a->kind == RMEM || a->kind == RCONST || (a->kind == RCPU && a->id != Esi && a->id != Edi);
	}

The UI8 op failed during a MOV8 (called by copy-copyfrom), with a RCPU id = 1, b RCPU id = 7 (Edi)

hashlink/src/jit.c

Lines 552 to 554 in 8bc5f8f

	case MOV8:
	if( !is_reg8(a) || !is_reg8(b) )
	ASSERT(0);

So I move the !is_reg8(from) also for the case when to.kind = RCPU. I also force the Reg choice to RCPU_8BITS because by the name it make sens, but it seems not necessary for the fix.

[ncannasse]

Seems good