Update URL config protocol for loading TF models

[reallsx]

Before you submit this PR: make sure to put all operations-related information in a wiki-note, a PR should be about code and is publicly accessible

What does the code in this PR do / what does it improve?

Make the loading of tensorflow model (for position reconstruction) compatible with the newer versions of tensorflow.

Can you briefly describe how it works?

1. If the file ends with .keras, load the file directly with tf.keras.models.load_model.
2. If the file ends with .tar.gz, extract the tarball. Then first execute the registration.py in it, then load the .keras file in it with tf.keras.models.load_model.

Can you give a minimal working example (or illustrate with a figure)?

Please include the following if applicable:

• Update the docstring(s)
• Update the documentation
• Tests to check the (new) code is working as desired.
• Does it solve one of the open issues on github?

Notes on testing

• Until the automated tests pass, please mark the PR as a draft.
• On the XENONnT fork we test with database access, on private forks there is no database access for security considerations.

All italic comments can be removed from this template.

[reallsx]

Seems like the tests failed because keras is not able to load the model...

While this PR is for changing the protocol for loading models for TF>2.18 which comes with keras 3 the testing environment is still using TF 2.15 so I guess this is why it is failing.

Collecting tensorflow==2.15.0.post1 (from base_environment==2024.1.1)

[dachengx]

Seems like the tests failed because keras is not able to load the model...

While this PR is for changing the protocol for loading models for TF>2.18 which comes with keras 3 the testing environment is still using TF 2.15 so I guess this is why it is failing.

Collecting tensorflow==2.15.0.post1 (from base_environment==2024.1.1)

We can try to update it to tensorflow>2.18, after

• ONLINE corrections migrated to xedocs https://github.com/XENONnT/corrections/pull/268
• CMT is removed Remove CMT #1475
• pdmongo package is not needed
• numpy is upgraded to >2

@LuisSanchez25 @rsriya @ghusheng @GiovanniVolta

[dachengx]

Is it normal to see warning

/home/xudc/.local/lib/python3.9/site-packages/keras/src/saving/saving_lib.py:757: UserWarning: Skipping variable loading for optimizer 'adam', because it has 18 variables whereas the saved optimizer has 2 variables. 
  saveable.load_own_variables(weights_store.get(inner_path))
<Sequential name=sequential, built=True>

when loading https://github.com/XENONnT/corrections/blob/3e8727006511081fc2cc6bd4c5948eba673d65d8/XENONnT/position_reconstruction/data/v9.json#L46C19-L46C47?

[reallsx]

Is it normal to see warning

/home/xudc/.local/lib/python3.9/site-packages/keras/src/saving/saving_lib.py:757: UserWarning: Skipping variable loading for optimizer 'adam', because it has 18 variables whereas the saved optimizer has 2 variables. 
  saveable.load_own_variables(weights_store.get(inner_path))
<Sequential name=sequential, built=True>

when loading https://github.com/XENONnT/corrections/blob/3e8727006511081fc2cc6bd4c5948eba673d65d8/XENONnT/position_reconstruction/data/v9.json#L46C19-L46C47?

This happens in my environment as well and the warning appears only once in a session. The warning is about the optimizer which has nothing to do when loading the model for inference. I think it is fine.

[coveralls]

coverage: 89.688% (-0.3%) from 89.966%
when pulling 4f3e41d on tf_protocal
into 977c2ea on master.