reduce excessive warnings and logging

[seungchan-an]

This PR reduces excessive warnings and logs during large CLM runs, which were generating very large Slurm log files.
All changes affect logging and warnings only and do not modify model behavior.

Changes

1. Suppress FutureWarning during train_models_RNN

• Before: FutureWarning (Series.__getitem__ treating keys as positions is deprecated ...) was emitted once per training sample.
• File: src/clm/datasets.py
• Descriptor values are converted to numpy before being passed to torch.Tensor.

2. Suppress SettingWithCopyWarning during create_training_sets

• Before: SettingWithCopyWarning (A value is trying to be set on a copy of a slice from a DataFrame ...) was emitted once per training sample.
• File: src/clm/commands/create_training_sets.py
• SMILES are explicitly converted to python lists before shuffling.

3. Make add_carbon logging optional

• Before: Progress messages (e.g. working on SMILES 501 of 1983: ...) were printed once per training sample.
• File: src/clm/commands/add_carbon.py
• A verbose flag was added so detailed progress output is printed only when requested.