Add saturating subtraction with a SIMD encoding

This includes the new instructions `ssub_sat` and `usub_sat` and only encodes the i8x16 and i16x8 types; these are what is needed for implementing the SIMD spec (see https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md#saturating-integer-subtraction).

cranelift/filetests/filetests/isa/x86/simd-arithmetic.clif

@@ -190,3 +190,31 @@ ebb0:
     return v4
 }
 ; run
+
+function %sub_sat_i8x16() -> b1 {
+ebb0:
+[-, %xmm2]    v0 = vconst.i8x16 [128 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0] ; 120 == 0x80 == -128
+[-, %xmm3]    v1 = vconst.i8x16 [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
+
+[-, %xmm2]    v2 = ssub_sat v0, v1 ; bin: 66 0f e8 d3
+    v3 = extractlane v2, 0
+    v4 = icmp_imm eq v3, 0x80 ; still -128, TODO it's unclear why I can't use -128 here
+
+    ; now re-use 0x80 as an unsigned 128
+[-, %xmm2]    v5 = usub_sat v0, v2 ; bin: 66 0f d8 d2
+    v6 = extractlane v5, 0
+    v7 = icmp_imm eq v6, 0
+
+    v8 = band v4, v7
+    return v8
+}
+; run
+
+function %sub_sat_i16x8() {
+ebb0:
+[-, %xmm3]    v0 = vconst.i16x8 [0 0 0 0 0 0 0 0]
+[-, %xmm5]    v1 = vconst.i16x8 [1 1 1 1 1 1 1 1]
+[-, %xmm3]    v2 = ssub_sat v0, v1 ; bin: 66 0f e9 dd
+[-, %xmm3]    v3 = usub_sat v0, v1 ; bin: 66 0f d9 dd
+    return
+}