Add optional ML-based solvation correction pathway

[BonhyeokKoo]

Summary

This PR introduces an optional ML-based solvation correction pathway into RMG’s thermochemistry pipeline. It provides an interface for estimating solvation effects using a pre-trained ML model, while preserving the existing LSER (Linear Solvation Energy Relationship) method as a fallback mechanism.

Motivation or Problem

Currently, RMG uses LSER for solvation corrections for thermo. At this stage, the ML component is implemented as a dummy version that always outputs zero corrections. This is intentional — actual ML functionality will be introduced later, after transitioning to Python 3.11 for better compatibility.

Description of Changes

Add MLSolvation class to rmgpy/data/solvation.py

• Introduced a new MLSolvation class that mirrors the structure of MLEstimator from mlEstimator.
• Accepts the same kind of mlSolvation block in the input file, with options like use_ml_solvation=True and name='solvation'.
• Unlike MLEstimator, which resides in rmgpy/ml/estimator.py, this new class is defined within rmgpy/data/solvation.py.
• The class exposes a get_solvation_correction() method, analogous to SolvationDatabase, and returns a SolvationCorrection object.

Add ml_solvation() function to rmgpy/rmg/input.py

• Mirrors the implementation of the ml_estimator() block and allows ML solvation to be optionally configured via the input file.

Modify rmgpy/thermo/thermoengine.py to support fallback logic

• At the point where solvation corrections are applied to thermo, the code now tries to use the ML-based model via get_input("ml_solvation").
• If this fails (e.g., due to configuration or import issues), the code gracefully falls back to the default LSER method.
• This behavior is wrapped in a try-except block.

Testing

• Two minimal examples were tested:
  1. Without an mlSolvation block → expected warning:

     Warning: ML solvation correction not used: 'RMG' object has no attribute 'ml_solvation'
     

     The system then falls back to LSER as expected.
  2. With an mlSolvation block → dummy ML model successfully invoked:

     [NOTICE] Dummy ML model loaded from: /Users/bon/rmg/RMG-database/input/thermo/ml/solvation
     [NOTICE] Dummy ML model utilized
     

Reviewer Tips

• Please check whether the following imports are appropriate and safe:
  • from rmgpy.data.solvation import SolvationCorrection in solvation.py
  • from rmgpy.rmg.input import get_input in thermoengine.py
• Note that MLSolvation is currently a dummy implementation and does not load an actual model. It is prepared to be upgraded in future commits after Python 3.11 migration.

[JacksonBurns]

hi @BonhyeokKoo welcome to RMG world! Please tag me in this PR as needed, and as a reviewer when the time comes.

Following up on some offline discussion, the following things need to happen to unblock this PR by adding support for Python 3.11:

• Merge Python 3.9 #2741 to unblock Python 3.9
• Complete and merge Use rdkit for SSSR and RCs (bug fix + Python upgrade) #2796 to remove our dependency on RingDecomposerLib which is incompatible with Python 3.11
• Complete and merge Update to cantera 3.x #2751 to add support for Python 3.11 (!)