Multiplication of shorts is promoted to multiplication of ints

The initial XLS IR output from xlscc for multiplication of `short`s first does a sign_ext to promote both inputs to `int`s, then multiplies them, then returns the lower 16 bits of the result. This makes the optimizer work harder than is necessary, and some targets might not be able to optimize it away.

```
#pragma hls_top
short foo(short a, short b){
return a*b;
}
```