Predict timesteps for ASAD in advance

Transfer of [ukca:#144](https://code.metoffice.gov.uk/trac/ukca/ticket/144) from MOSRS.

The ASAD chemistry solver currently includes redundant calculation. For each chunk of data passed to the solver, if any of the corresponding grid-boxes fail to solve then the entire chunk is solved again at half the timestep until convergence is achieved. If we were able to predict the required timestep in advance then we could avoid this redundant calculation.

Predicting the timestep in advance is not trivial, however. This is ongoing research work on whether we are able to map data available to the solver at its initialisation to the eventual number halving steps and/or required timestep. The process may or may not include machine learning.

Start by examining the ASAD chemistry solver and the data available during solver initialization, then trace how failed grid-boxes trigger chunk retries and timestep halving. Review the transferred MOSRS ticket for context. Done would require evidence that the required timestep or number of halving steps can be predicted reliably, but the issue does not define a specific implementation or acceptance test.

Written by the indexing model from the issue text.