Fix clipping float dtypes when either min or max are Python ints

[magnusdk]

This PR makes clip no longer crash when x has a floating point dtype and min and/or max are Python ints.

import array_api_compat.numpy as xp
xp.clip(xp.asarray(0.0), 0, 1)  # <- this used to crash

The Array API for clip states that behavior is unspecified if xand either min or max have different data type kinds, but I think the change in this PR makes it behave more like what one might expect. This change also makes the code more similar to NumPy's handling of Python ints in clip.

This is my first PR for this repo, hopefully I got it right :) Let me know if I should add anything else.
The work you do with array-api is great, thanks a lot!

[ev-br]

I checked locally that indeed, the behavior of numpy, pytorch, jax.numpy and dask.array is to allow xp.clip(xp.asarray(0.0), 0, 1) and return a floating-point 0.0.
The wrapped namespaces all currently reject this invocation.

The behavior is indeed unspecified in the standard, and is thus implementation-defined. array-api-compat by design should not disallow implementation-defined behaviors.

The itself change looks safe.

Therefore, LGTM.

Let's merge this patch, thank you @magnusdk!