Add logging to file with MLflow integration

[loren-ac]

Description

Currently, logs only output to stdout. We should add the ability to log to a file
and automatically save that log file to MLflow for better debugging and analysis.

Context

When runs encounter errors or need debugging, having a complete log file saved in
MLflow would be valuable for post-mortem analysis and tracking training history.

Requirements

1. Configure Python's logging library to write to both stdout and a file

   • Research the logging library's configuration for multiple handlers
   • Add a file handler in addition to the existing stream handler

2. Add log file path configuration

   • Option 1: Add a log_file_path parameter to the run configuration
   • Option 2: Use a default path convention (e.g.,
     logs/{experiment_name}/{run_name}.log)

3. Integrate with simplexity run manager

   • Ensure log file is saved to MLflow in the cleanup/teardown phase
   • Include log file upload even when runs fail/error out
   • Log file should be available as an MLflow artifact

Acceptance Criteria

• Logs are written to both stdout and a file simultaneously
• Log file path is configurable (either via config or using sensible defaults)
• Log files are automatically uploaded to MLflow as artifacts
• Log files are captured and uploaded even when runs fail with errors
• Documentation updated to explain log file configuration options

Notes

The log file path needs to be known by the run manager's cleanup function to ensure
upload happens reliably.