next: np.bool_ argument handling broken in gtfn backend

[nfarabullini]

The following test outputs an unexpected error:

@pytest.mark.datatest
def test_bool_fo(icon_grid, backend):
    import gt4py.next as gtx
    from icon4py.model.common.utils import data_allocation
    from icon4py.model.common import field_type_aliases as fa
    from gt4py.next.ffront.fbuiltins import broadcast
    import numpy as np
    @gtx.field_operator
    def bool_fo(val: bool) -> fa.EdgeField[bool]:
        return broadcast(val, (EdgeDim,))
    val = np.bool_(True)
    f = data_allocation.zero_field(icon_grid, dims.EdgeDim, dtype=bool)
    bool_fo.with_backend(backend)(val, out=f, offset_provider={})

Error:

>       bool_fo.with_backend(backend)(val, out=f, offset_provider={})

dycore_tests/test_velocity_advection.py:474: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../../../.venv/lib/python3.10/site-packages/gt4py/next/ffront/decorator.py:604: in __call__
    return self.backend(
../../../../.venv/lib/python3.10/site-packages/gt4py/next/backend.py:157: in __call__
    self.jit(program, *args, **kwargs)(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

offset_provider = {}
out = NumPyArrayField(_domain=Domain(dims=(Dimension(value='Edge', kind=<DimensionKind.HORIZONTAL: 'horizontal'>),), ranges=(UnitRange(0, 31558),)), _ndarray=array([False, False, False, ..., False, False, False]))
args = (False, NumPyArrayField(_domain=Domain(dims=(Dimension(value='Edge', kind=<DimensionKind.HORIZONTAL: 'horizontal'>),), ranges=(UnitRange(0, 31558),)), _ndarray=array([False, False, False, ..., False, False, False])))
converted_args = [False, (array([False, False, False, ..., False, False, False]), (0,))]

    def decorated_program(
        *args: Any,
        offset_provider: dict[str, common.Connectivity | common.Dimension],
        out: Any = None,
    ) -> None:
        if out is not None:
            args = (*args, out)
        converted_args = [convert_arg(arg) for arg in args]
        conn_args = extract_connectivity_args(offset_provider, device)
        # generate implicit domain size arguments only if necessary, using `iter_size_args()`
>       return inp(
            *converted_args,
            *(arguments.iter_size_args(args) if inp.implicit_domain else ()),
            *conn_args,
        )
E       TypeError: __field_operator_bool_fo(): incompatible function arguments. The following argument types are supported:
E           1. __field_operator_bool_fo(arg0: bool, arg1: tuple[ndarray[dtype=bool, shape=(*)], tuple[int]], arg2: int, /) -> None
E       
E       Invoked with types: bool_, tuple, int

../../../../.venv/lib/python3.10/site-packages/gt4py/next/program_processors/runners/gtfn.py:59: TypeError

[havogt]

nanobind lacks support for casting np.bool to C++ bool. See wjakob/nanobind#936.

[havogt]

Since most likely nanobind doesn't want the ugly np.bool cast, we should just wrap bool args with a bool() cast.