Centralise dataset keys and run types in a config file

[jeipollack]

Problem:
Data extraction across modules (data_handler.py, centroids.py, ZernikeInputsFactory, etc.) currently uses hard-coded keys and run-type logic (i.e. "simulation", "training", "metrics" and "inference"). This leads to duplicated code, scattered special cases, lengthy conditionals and makes adding new run types or datasets harder.

In PR #163, I am introducing get_data_array with helper _get_direct_data consolidates handling of these cases, but further improvements are possible to enable a cleaner, more abstract design.

Proposal:

• Introduce a config file to store:
  • Dataset keys for each data type (e.g., positions, stamps, masks).
  • Run-type properties (e.g., whether to concatenate train/test datasets).
  • Refactor get_data_array and related dataset utilities to read from this config instead of using hard-coded keys.

Benefits:

• Centralised abstraction for keys and run-type logic.
• Cleaner, leaner module code.
• Easier extension for new run types or real observational data.