Warmstart infrastructure switch

[le1nux]

What does this PR do?

This PR mainly addresses the warmstart of model training, e.g., after GPU crashes.

General Changes

• Fixes issue Warmstart checkpoints have correct amount of steps but wrong amount of tokens #242
• Warmstarts with changing infrastructure (e.g.,. different number of GPUs) are now supported.
• Restructures the settings part of the configs to
• Adds various checks for consistency of model training (e.g., target tokens and number of dataset tokens mismatch)
• Refactors all configs to be runnable again
• Adds an interactive jupyter notebook-based Tutorial on how to use Modalities. (merged from PR Interactive getting started tutorial #239 )
• Adds a warmstart tutorial
• TrainingReportGenerator that creates a report on the training setup and prints out warnings in case of inconsistencies.
• Activation Checkpointing is now a component
• Added further NumberConversion routines

Breaking Changes

• the settings part of the configs have been completely refactored

Checklist before submitting final PR

• My PR is minimal and addresses one issue in isolation
• I have merged the latest version of the target branch into this feature branch
• I have reviewed my own code w.r.t. correct implementation, missing type hints, proper documentation, etc.
• I have run a sample config for model training
• I have checked that all tests run through (python tests/tests.py)
• I have updated the internal changelog (CHANGELOG_DEV.md)

[flxst]

Added a few comments mainly about the modalities in 15 mins tutorial and README.md

[flxst]

Would still be nice to change this to "Multimodal Foundation Model Training" (without the "s")

[fromm-m]

looks good to me

[flxst]

LGTM :)