[ENH] Fix model caching at refit and checkpointing in v2

[Faakhir30]

Reference Issues/PRs

Fixes #2338

What does this implement/fix? Explain your changes.

• disable model caching completly at fit
• fix checkpoint loading to ensure that checkpointed metadata is same as passed metadata
• added tests for changes

Other Alternative

• dont raise an error, instead warn and recreate the model with new metadata

Did you add any tests for the change?

Yes

PR checklist

• The PR title starts with either [ENH], [MNT], [DOC], or [BUG]. [BUG] - bugfix, [MNT] - CI, test framework, [ENH] - adding or improving code, [DOC] - writing or improving documentation or docstrings.
• Added/modified tests
• Used pre-commit hooks when committing to ensure that code is compliant with hooks. Install hooks with pre-commit install.
  To run hooks independent of commit, execute pre-commit run --all-files

[codecov]

Codecov Report

❌ Patch coverage is 89.63585% with 37 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@e7c7965). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
pytorch_forecasting/base/_base_pkg.py	87.76%	17 Missing ⚠️
pytorch_forecasting/callbacks/artifact_registry.py	84.12%	10 Missing ⚠️
...forecasting/data/data_module/_tslib_data_module.py	69.56%	7 Missing ⚠️
...g/data/data_module/_encoder_decoder_data_module.py	90.00%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2350   +/-   ##
=======================================
  Coverage        ?   87.31%           
=======================================
  Files           ?      172           
  Lines           ?    10345           
  Branches        ?        0           
=======================================
  Hits            ?     9033           
  Misses          ?     1312           
  Partials        ?        0           

Flag	Coverage Δ
cpu	87.31% <89.63%> (?)
pytest	87.31% <89.63%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[phoeenniixx]

Will it make sense to stack this on #2323 ?

[Faakhir30]

Will it make sense to stack this on #2323 ?

Only thing overlapping is just different place to read checkpoint metadata, concept of rebuilding model/validating metadata for checkpoint will stay the same. afaik

While solving merge conflicts: we will have to read from datamodule_metadata path instead of Path(self.ckpt_path).parent / "metadata.pkl"

[phoeenniixx]

Will it make sense to stack this on #2323 ?

Only thing overlapping is just different place to read checkpoint metadata, concept of rebuilding model/validating metadata for checkpoint will stay the same. afaik

While solving merge conflicts: we will have to read from datamodule_metadata path instead of Path(self.ckpt_path).parent / "metadata.pkl"

I see, but still I think we should stack these PRs, this will help me review as well - based on the new save and load interface and not on the last one :)

[Faakhir30]

• stacked
• reimplemented
• moved self.metadata = metadata at BaseModel. Some models were not doing this, and reading directly from metadata passed in param.
• Failing test Test / Run pytest (macos-latest, 3.14) (pull_request) is internet connection issue