Add Windows x64 ABI support for amd64 target

[itsfuad]

• Introduced a new target structure for Windows x64 calling convention in amd64/targ.c.
• Updated main.c to include the new Windows x64 target in the target mapping.
• Implemented the Windows x64 ABI logic in amd64/win64.c, including argument classification, return value handling, and stack management.
• Created a dedicated emitter for Windows x64 in amd64/win64_emit.c, mirroring existing functionality while adhering to the Windows calling conventions.
• Added necessary utility functions for argument handling and register management specific to the Windows ABI.

Closes #4

[itsfuad]

The block-entry remap step in rega.c can assign a live temp to a register already used by another live temp in the same block.

• On amd64, this can break div/rem loops where two phi values are live at the loop header, causing the dividend and divisor to share a register (wrong results or infinite loops).

Repro (QBE IL):

export function w $gcd(w %a, w %b) {
@b1
	%x =w copy %a
	%y =w copy %b
	jmp @b2
@b2
	%y2 =w phi @b1 %y, @b3 %y3
	%x2 =w phi @b1 %x, @b3 %x3
	%c =w cnew %y2, 0
	jnz %c, @b3, @b4
@b3
	%tmp =w copy %y2
	%rem =w rem %x2, %y2
	%y3 =w copy %rem
	%x3 =w copy %tmp
	jmp @b2
@b4
	ret %x2
}

Before patch, qbe -dR shows %x2 and %y2 mapped to the same register at @b2, which makes idiv use the same reg for dividend/divisor. After patch, they stay in distinct registers.

Fix:

• In rega.c step “emit copies shared by multiple edges,” skip remapping a temp to a register that is already assigned to another live temp in the block.
• Only update m->r and the bitset when the target register is free.