BUG: torch.maximum: does not accept `float` for `x1`

[lucascolley]

scipy/sparse/linalg/_isolve/iterative.py:25: in _get_atol_rtol
    atol = xp.maximum(atol, rtol * b_norm)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        atol       = 0.0
        b_norm     = tensor([], size=(0, 1), dtype=torch.float32)
        name       = 'cg'
        rtol       = 0.0015
        xp         = <module 'scipy._external.array_api_compat.torch' from '/Users/lucascolley/ghq/github.com/scipy/scipy/build-cpu-install/usr/lib/python3.13/site-packages/scipy/_external/array_api_compat/torch/__init__.py'>
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

x1 = 0.0, x2 = tensor([], size=(0, 1), dtype=torch.float32), kwargs = {}

    @_wraps(f)
    def _f(x1, x2, /, **kwargs):
        x1, x2 = _fix_promotion(x1, x2)
>       return f(x1, x2, **kwargs)
               ^^^^^^^^^^^^^^^^^^^
E       TypeError: maximum(): argument 'input' (position 1) must be Tensor, not float

This is with array-api-compat at the vendored 1.14 tag in SciPy.

This contradicts https://data-apis.org/array-api/2025.12/API_specification/generated/array_api.maximum.html#maximum.

[lucascolley]

cc @ev-br

[ev-br]

I think it's #271

[lucascolley]

Ah yes, well bump on that!