Fix tests for upgraded dependencies

[stes]

Update: The root issue was numpy 2. This is a recent upgrade (June 2024), and it seems like many of the packages required for cebra do not easily work with npy2 (yet). Errors included pandas/pytables, but also torch. It would be probably possible to work around these errors, but I think it is not worth yet --- in half a year or so, the situation is probably more stable, and we can remove the pinning.

This PR now does the following changes:

• Pin numpy below v2.
• Remove nlb_tools to allow use of more recent pandas versions. This will only affect users who want to use the monkey dataset. We provide guidance on how to post-hoc install the package directly via a warning for these users.
• Fix a minor issue in the plotly code, which arises because we now use more recent versions of plotly.

I propose to merge as-is. From the user perspective, it is slightly annoying that CEBRA does not run with numpy 2 for now, but given the errors caused in other upstream packages, chances are that no one uses numpy 2 (yet) in the first place. We just still aim to remove the pinning again as soon as possible.

---

Old description:

This PR aims to solve some of the broken tests we saw in recent PRs (e.g. #166 #183 #175 #180 ). My hypothesis is that the nlb_tools package which we require for the test workflow right now is the root cause, which cases pandas to be downgraded to 1.3.4 (https://github.com/neurallatents/nlb_tools/blob/main/requirements.txt#L1), while the current pandas version is 2.2.3 (https://pypi.org/project/pandas/).

The build issue was a pandas <> numpy version mismatch (numpy was >2, pandas at 1.3.4 before), so hopefully this fixes the build.

I removed this depedency which causes more recent versions of all packages to be installed. This caused some plotly tests to break, which now do stricter checking on the colormaps. I also updated one of the helper functions there.

Let's see if this passes as-is on github, local tests look good.

[stes]

relevant packages in failing build:

h5py-3.12.1 (latest: 3.12.1)
pandas-2.2.3 (latest: 2.2.3)
numpy-2.0.2 (latest: 2.1.2)

[stes]

Our particular error was also discussed here, it seems like some of the packages (pandas, pytables, ...) simply do not work yet with numpy 2 which was released this june. testing what happens if we include this as a constraint.

[stes]

Related issue from the pytorch docs: pytorch/pytorch#107302 (seems merged though)

[stes]

@MMathisLab this is ready for review, and can be merged from my end. I updated the PR description with my line of thinking, if you agree feel free to merge.

[MMathisLab]

Thanks so much @stes