Should we add parameter checks in numba RV functions?

### Description

Not limited to, but Normal is an illustrative example:

```python
import pytensor.tensor as pt

x = pt.random.normal(0, -1)
x.eval(mode="NUMBA")
try:
    x.eval(mode="FAST_COMPILE")
except ValueError as e:
    print(e.args[0].split("\n")[0])  # scale < 0
```

It will probably slow down the impl