tracing: track query sent to prometheus via remote read api

[igorwwwwwwwwwwwwwwwwwwww]

This patch extends tracing to include a prometheus.query tag for queries against the prometheus remote read API (/api/v1/read).

Previously there wasn't much information about the call being made, since that API uses protobuf over POST bodies. Now there is a JSON encoding of the protobuf request body in the trace.

One potential downside of this change is that there is JSON encoding and allocation overhead on every request. I wasn't able to find an easy way to make those conditional (input very welcome).

I also moved spanReqDo.Finish() before the error handling. I believe we want to finish the span in any case -- though if leaving it open was intentional, I'm happy to change it back.

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Changes

• tracing: Add prometheus.query tag to query_prometheus_request span

Verification

I ran thanos locally with tracing configured for elastic APM and confirmed the addition of the prometheus.query tag. Example value:

{"start_timestamp_ms":1581930268943,"end_timestamp_ms":1581934168943,"matchers":[{"name":"__name__","value":"up"}]}

[bwplotka]

I think marshalling overhead is super small here, so happy with this 👍

LGTM, thanks!