Use regalloc constraints for sse blend operations (#5251)

Instead of using xmm0 explicitly for the mask argument to instructions like blendvpd, use regalloc constraints to constrain it to xmm0 instead.

cranelift/codegen/src/isa/x64/lower/isle.rs

@@ -336,11 +336,6 @@ impl Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6> {
         0b00_00_00_00 | lane << 4
     }
 
-    #[inline]
-    fn xmm0(&mut self) -> WritableXmm {
-        WritableXmm::from_reg(Xmm::new(regs::xmm0()).unwrap())
-    }
-
     #[inline]
     fn synthetic_amode_to_reg_mem(&mut self, addr: &SyntheticAmode) -> RegMem {
         RegMem::mem(addr.clone())