[Metrics] Collect more fine grained metrics for p2p comms

[jorgeantonio21]

In order for clients to best route their requests in the best possible way, it is necessary to collect the more fine grained metrics available:

• Accelerator max and optimal thermal and power values (vendor and accelerator specific);
• Number of received inference requests for given model;
• Inter token latency (on average);
• Time to first token (on average);
• Number of total requests;
• Number of successful requests;
• Number of failed requests.

[maschad]

I've removed the first two metrics for a separate issue as it's more involved