Use 32-bit XORRegReg to zero registers on x64

[0xdaryl]

When XORRegReg is used to zero a register on x64, the 32-bit operand form should be used because the upper 32-bits of the result are automatically zero extended. This will save a REX prefix on each of these instructions (1 byte savings).

Sweep through the x86 code generator looking at where XORRegReg or XOR8RegReg is used for the purpose of zeroing a register and ensure it uses XOR4RegReg instead. I found a few on inspection.