Fix + add argument to pad for pytorch backend

[FlorianPfaff]

Checklist

• My pull request has a clear and explanatory title.
• If necessary, my code is vectorized.
• I added appropriate unit tests.
• I made sure the code passes all unit tests. No, but I'm not sure if that might be some random failure?
• My PR follows PEP8 guidelines. (refer to comment below)
• My PR follows geomstats coding style and API.
• My code is properly documented and I made sure the documentation renders properly. (Link)
• I linked to issues and PRs that are relevant to this PR.

Description

In the pytroch backend, the argument used for padding values was "constant_value". However, the naming in jax and numpy is constant_values. See https://numpy.org/doc/stable/reference/generated/numpy.pad.html
and https://jax.readthedocs.io/en/latest/_autosummary/jax.numpy.pad.html. I also added passing on the "mode" argument, which is supported by pytorch.

Issue

This patch ensures that passing "constant_values" is compatible with all backends.

Additional context

[luisfpereira]

Thank you @FlorianPfaff. Merging. (Sorry for taking so long.)