feat(agent): Add leeching throughput metric

[Anton-Kalpakchiev]

Replace agent's time_per_mb latency metric with a p2p_leech_throughput metric in MiB/s.

Benefits:

• we now emit metrics on <1MB blobs - this includes a SIGNIFICANT chunk of kraken download blob requests, as every image has 1-3 image manifests, which are <1MB.
• we use a histogram instead of a timer (lower cardinality)
• all throughput metrics are now centralized under lib/observability
• we now tag blobs by size when emitting the leeching throughput
• I also added a few more throughput buckets, in case throughput is very high (also since I'll be reusing this function for the origin downloads from GCS)
• Finally, I added a safeguard in the EmitDownloadPerformance function against a user providing latency=0, which would result in a panic when emitting the mbPerSec (some unit tests use clock.NewMock and unfortunately can't increment the clock in the middle of a download).