Fix checkpoint resume for Monte Carlo GCS

[mmaclay]

Problem

The checkpoint resume capability for Monte Carlo GCS currently does not preserve nadir-equivalent error calculations, since nadir-equivalent error is calculated in the subsequent error-stats step. This prevents accurate aggregation of error statistics across multiple runs since the required error data isn't saved in the NetCDF files.

Proposed Improvements

1. Calculate and Write Nadir-Equivalent Error:
   • Implement or port the nadir-equivalent error calculation logic from geolocation_error_stats.py into image_match.py.
   • Ensure this error statistic is computed for each checkpoint and written to the NetCDF output during simulation.
2. Loop Integration:
   • Integrate the calculation step directly into the main simulation loop so checkpoints will have up-to-date NE error values.
   • At the end of the simulation batch, run the full error statistics aggregation for all completed and resumed runs.
3. Support Checkpoint Resume:
   • When resuming from a checkpoint, read previously saved nadir-equivalent errors from NetCDF to incorporate them into aggregated statistics.
   • Ensure resume behavior is robust for both partial and full run aggregation.

Implementation Plan

• Audit existing error metrics code in geolocation_error_stats.py and image_match.py to identify what needs porting or integration.
• Refactor the code so nadir-equivalent error calculation is available and called within the simulation loop where checkpoints are written.
• Update NetCDF output handling to include nadir-equivalent error for each checkpoint.
• Develop logic to load and aggregate previous checkpoint errors when resuming simulations.
• Add/modify unit tests to cover resume and aggregation of error statistics.
• Document the new error calculation workflow and update relevant usage guides.

---

Background:
The nadir-equivalent error metric is essential in post-processing, allowing direct inclusion of previous runs' error values in aggregate statistics. Integrating it directly with periodic checkpointing will improve both robustness and downstream error analysis.

---

Additional Notes:

• Review whether any non-obvious dependencies between error calculation and NetCDF writing need explicit handling.
• Confirm compatibility for current checkpoint format and resume workflow.

---

Original issue below for context.

---

We need nadir-equivalent error to be calculated and written to netcdf in order to be able to resume and include previous runs in error stats calculations.

We should explore porting the nadir-equivalent error calculation step from geolocation_error_stats.py to the image_match.py or at least call that calculation IN loop, and THEN run the full error Stats at the end of the loop (for aggregate stats).