Add ResNet-based encoder, Perceptual loss, KLAnnealing, and ParamManager

[BerndDoser]

• Added a ResNet-based encoder, an upsampling decoder
• Added a Perceptual loss
• Added a new KLAnnealing callback (src/spherinator/callbacks/kl_annealing.py) for flexible annealing of the KL divergence weight during VAE training
• Added a ParamManager callback (src/spherinator/callbacks/param_manager.py) for parameter value management and freezing based on regex patterns
• Improved LogReconstructionCallback to handle datasets with augmentations, logging original, augmented, and reconstructed images in a single figure
• Updated .github/workflows/python-package.yml to use the latest versions of GitHub Actions and setup-uv, improving CI reliability and compatibility
• Added ruff checks for linting and formatting to .github/check.sh for improved code quality enforcement