Short uniform vector real size

It looks like the short vector size is actually rounded up to the next power of 2. It is probably done for small sizes due to some performance concerns (e.g., 3 to 4), despite the fact that it is error prone (easy to write an out-of-bound read/write). If short vector has bigger size then it also rounded up. For the example below (https://godbolt.org/z/1P7TE961o), the real size if 512 instead of expected 400.

```ispc
uniform int foo() {
return sizeof(uniform float<100>);
}
```

```asm
foo___: # @foo___
mov eax, 512
ret
```

Rounding is implemented here https://github.com/ispc/ispc/blob/80dccf9c2ff614cb5e144250757e80b1d2263f3e/src/type.cpp#L1992-L2004