Get progress metrics in a file during run, not after the result is completed

[rishabh-nuta]

Is your feature request related to a problem? Please describe.
GuideLLM currently writes benchmark metrics only after the run completes. During long-running benchmarks, progress metrics are available only via the interactive TTY UI, which is not usable in automated or non-TTY environments

Describe the solution you'd like
Add support for emitting progress metrics to a file periodically during the run (e.g., incremental JSON/JSONL output flushed at --output-sampling intervals), so progress can be monitored programmatically without requiring a TTY.

Additional context
This would significantly improve observability and automation for long-running benchmarks and production test pipelines.

[rishabh-nuta]

@sjmonson @markurtz can you please help us here, or maybe guide us for some patchwork

[sjmonson]

The live metrics output to the console contains only a few counters and averages. If that is all you want then it should not be too hard to generate some periodic log messages. We are looking to improve logging especially in non-interactive ttys Generating the full metrics report without affecting benchmark performance could be a challenge so I would focus on the former first.

[sjmonson]

To answer your second question a little better. The main benchmarking loop is here. You can look at the accumulator and the methods that take it as an argument for logging status.