add response duration metric historgram with buckets

[jakubgs]

This way we can measure spread of latencies across multiple requests over time and not just duration time of the last request made.

Example:

# HELP presto_server_prepare_response_duration_seconds Histogram of HTTP response preparation duration, in seconds
# TYPE presto_server_prepare_response_duration_seconds histogram
presto_server_prepare_response_duration_seconds_sum{endpoint="GET/eth/v1/beacon/states/{state_id}/fork"} 1.028
presto_server_prepare_response_duration_seconds_count{endpoint="GET/eth/v1/beacon/states/{state_id}/fork"} 6.0
presto_server_prepare_response_duration_seconds_created{endpoint="GET/eth/v1/beacon/states/{state_id}/fork"} 1777402883.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.01"} 4.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.025"} 4.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.05"} 4.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.1"} 4.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.25"} 4.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="0.5"} 5.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="1.0"} 6.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="2.5"} 6.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="5.0"} 6.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="10.0"} 6.0
presto_server_prepare_response_duration_seconds_bucket{endpoint="GET/eth/v1/beacon/states/{state_id}/fork",le="+Inf"} 6.0

Personally I think presto_server_prepare_response_time metric is almost useless because high latency responses can be easily lost because the next set() call can reset the value back to a normal looking one.

Removing it entirely might be even better. Happy to do that if people agree.

[jakubgs]

I'm not sure how detailed we want the buckets, here's a more fine-grained example:

Not sure how valuable is the distinction between 1 ms and 10 ms.