Update jnp.floor_divide to make it consistent with np.floor_divide for division by zero

[rajasekharporeddy]

Fixes #11990

Current behaviour:

>>> jnp.floor_divide(2, 0)
Array(-2, dtype=int32, weak_type=True)
>>> np.floor_divide(2, 0)
<stdin>:1: RuntimeWarning: divide by zero encountered in floor_divide
0

With thi PR:

>>> jnp.floor_divide(2, 0)
Array(0, dtype=int32, weak_type=True)

[jakevdp]

Using _where would make this more succinct. I think this should work:

Suggested change

	result = lax.div(x1, x2)
	result = lax.select(_broadcast_to(x2 == 0, result.shape),
	lax.full_like(result, fill_value=0),
	result)
	return result
	return _where(x2 == 0, 0, lax.div(x1, x2))

[jakevdp]

Similarly here:

Suggested change

	result = _where(select, quotient - 1, quotient)
	result = lax.select(_broadcast_to(x2 == 0, result.shape),
	lax.full_like(result, fill_value=0),
	result)
	return result
	return _where(x2 == 0, 0, _where(select, quotient - 1, quotient)

[jakevdp]

We need to double check this logic for when x2 is -0.0 (the sign of the returned infinity may be wrong), or when x1 and x2 are both zero (when the result should be NaN).