Cranelift: Saturating arithmetic with SIMD vectors with lane-widths >16 bits not implemented in x86_64 backend

Opening an issue based on discussion in https://github.com/bytecodealliance/wasmtime/pull/3188#discussion_r690657131.

`uadd_sat` and `usub_sat` are not implemented in the `x86_64` Cranelift backend (but are implemented for `AArch64`) for `i32x4` or `i64x2` SIMD vectors.

CLIF bugpoint (`uadd_sat`):
```
Crash message: Unsupported type for packed uadd_sat instruction: i32x4

function %uaddsat_i32x4() -> i32x4 system_v {
const0 = 0x00000000000000000000000000000000

block0:
v0 = vconst.i32x4 const0
v1 = vconst.i32x4 const0
v2 = uadd_sat v0, v1
return v2
}

Crash message: Unsupported type for packed uadd_sat instruction: i64x2

function %uaddsat_i64x2() -> i64x2 system_v {
const0 = 0x00000000000000000000000000000000

block0:
v0 = vconst.i64x2 const0
v1 = vconst.i64x2 const0
v2 = uadd_sat v0, v1
return v2
}
```

CLIF bugpoint (`usub_sat`):
```
Crash message: Unsupported type for packed usub_sat instruction: i32x4

function %usubsat_i32x4() -> i32x4 system_v {
const0 = 0x00000000000000000000000000000000

block0:
v0 = vconst.i32x4 const0
v1 = vconst.i32x4 const0
v2 = usub_sat v0, v1
return v2
}

Crash message: Unsupported type for packed usub_sat instruction: i64x2

function %usubsat_i64x2() -> i64x2 system_v {
const0 = 0x00000000000000000000000000000000

block0:
v0 = vconst.i64x2 const0
v1 = vconst.i64x2 const0
v2 = usub_sat v0, v1
return v2
}
```