Cpf fails with torch due to slicing with negative strides

[sandorkertesz]

What happened?

test_cpf_core fails with this input/options when the array format is torch:

(_cpf.cpf_clim3, _cpf.cpf_ens3, dict(sort_clim=True, symmetric=True), _cpf.cpf_val3), # sym

        if symmetric:
>           cpf_reverse = _cpf(-clim[::-1, :], -ens[::-1, :], from_zero=from_zero)
                                ^^^^^^^^^^^^^
E           ValueError: step must be greater than zero

src/earthkit/meteo/extreme/array/cpf.py:150: ValueError

The reason is that slicing with negative strides is not supported in torch.

What are the steps to reproduce the bug?

See above

Version

0.5.0

Platform (OS and architecture)

Any

Relevant log output

Accompanying data

No response

Organisation

ECMWF

[Oisin-M]

Came across this in #74 and a fix is to use xp.flip(array, axis=0), which is part of the array-api-spec and supported by torch, cupy and numpy. Caveat: torch.flip returns a copy, np.flip returns a view.