Collect telemetry metrics from Triton metrics endpoint

[lkomali]

This PR is a part of adding telemetry metrics to GenAI-perf.

Changes implemented in this PR:

1. Added a new Telemetry metrics class that has the GPU metrics mentioned below.
   • gpu_power_usage
   • gpu_power_limit
   • energy_consumption
   • gpu_utilization
   • total_gpu_memory
   • gpu_memory_used
2. A new folder to hold all telemetry data collection files
3. A Base Telemetry Data Collector class that implements a thread to gather metrics from endpoint.
4. A specific Triton Telemetry Data Collector class that has a specific implementation for Triton metrics.
5. Modified wrapper.py to start TelemtryDataCollector thread once perf_analyzer subprocess starts running.

This is how the metrics are stored.

For every metric, the values are stored as a list of lists.
The outer list represents the sequence of metric measurements over time.
Each inner list contains the metric values for each GPU at a particular time point.

[debermudez]

Nice job with this.
I want to go through some of it with you but this is a great step towards wrapping this up and getting this out to our users. Thanks for working so hard on this.

[dyastremsky]

Excellent work, Harshini!

[debermudez]

Just wait to merge until @nv-hwoo closes his comments.
Nice work!

[nv-hwoo]

Awesome work @lkomali! Thanks for adding such a detailed doc string 👍 And sorry for getting back so late 😅 All looks good with just one small comment about the optional argument for TelemetryDataCollector in the run function.