Add nnunet one-line trainer (DAT-897)

[luandalmazo]

Adds NNUNetTrainer, which runs the full nnUNet v2 pipeline (export -> fingerprint -> plan -> preprocess -> train -> predict -> import) in a single trainer.fit() call, logging per-epoch metrics to MLflow. Also adds the tutorial notebook.

Important to note:
The deploy bundle patches trainer_name = 'nnUNetTrainer' inside checkpoint_final.pth before logging to MLflow. This is basically a stopgap for DAT-903 (the deploy container cannot install datamint from PyPI when running a dev version). The patch is safe as long as _DatamintNNUNetTrainer does not override the network architecture. When we solve DAT-903 we should remove this patch and set nnUNet_extTrainer in load context, so users can modify the network if they want.

Closes DAT-897.

[Lucashsmello]

I'll trust you on this PR. :)

Check test results with python3.10: https://github.com/SonanceAI/datamint-python-api/actions/runs/27279659681/job/80570314031?pr=67
I guess you used a feature of f-string not available on Python 3.10

[luandalmazo]

All reviewed and tested. I performed both unit and integration tests multiple times. Currently, the Git CI tests are returning errors due to the specific Python version being used and its particularities. Because of this, I have opened a new task (DAT-936) to add fallbacks and fix the tests.