When logging execution times, include more information in the log record

[bzorn]

The current timing info for an eval looks like this:

llm.eval.test: 12748ms

It would be great to also include:

• the prompt being tested
• the test number
• the LLM being tested

[bzorn]

Here is a suggested format for the timing data.
Create a csv file called timing.csv in the directory with the execution trace with the following fields:

action, sample name, baseline model, gen model, eval model, model under test, time (ms)
gen.intent, "speech-tag", "gpt4o","gpt4o", "gpt4o", "phi4", 1356
test.run, "speech-tag", "gpt4o","gpt4o","gpt4o", "phi4", 125132

etc.

[bzorn]

Will add back if more detailed performance analysis is needed.