Add compression comparison functionality

[BradleyBooth]

Added

• New autoencoder model in models.py - AE_float32
  When Baler compressed float32 data using the default AE model, it resulted in compressed files larger than the original. This was due to the layers all being hardcoded to float64. Using this model with float32 data avoids the issue.

  • New model inherits from existing AE model class.
  • Linear layers are modified to use dtype=torch.float32
  • To utilise the model add c.float_dtype = "float32" to the project config file.

• Lossy Compression Comparison functionality (compare.py)
  New baler operating mode defined in baler.py to benchmark baler performance on the current project against a selection of lossy compression approaches

  • To access, run baler using --mode compare

[jlsmith-hep]

Suggested change

	green_code_tracking.txt
	*.txt
	*.npz
	*.dat
	*.png
	*.root
	*.jpg
	*.jpeg
	*.log

Are there any .txt files we would want to track? Maybe a catch-all for results/data/log files would be better

[jlsmith-hep]

Are we sure project_name is always provided? Do we have a default? Just thinking this would break compatibility with an old script if there isn't a default, and this isn't defined

[jlsmith-hep]

Suggested change

	# Copyright 2022 Baler Contributors
	# Copyright 2022-2025 Baler Contributors

And similar for other files

[jlsmith-hep]

Hi @BradleyBooth , nice PR! Left a couple small comments, but I have a bigger one - there's a lot of refactoring into new methods and classes - are we sure this doesn't affect functionality? Is there any validation to look at, e.g. running a bundled example like CMS or CFD and seeing that they give the same output?