Add metrics about Matrix API calls

[jaller94]

This is an alternative to #67.

This adds the following Prometheus metrics:

• bridge_matrix_api_calls Counter: Number of API calls made
• bridge_matrix_api_calls_failed Counter: Number of API calls which failed

The names and types have been chosen to match those of https://github.com/matrix-org/matrix-appservice-bridge.

Incomplete list of things which could be improved

• Add a meaningful method labels to other parts of the API, not just events.py.

Differences to PR #67

• This covers all Matrix API calls, not just those in events.js, however, all others have an empty string as the method label.
• This counts retries as separate API calls. Add metrics about Matrix API events #67 considers retries to be part of the first call.
• Removed bridge_matrix_request_seconds s it was implemented differently from the NodeJS bridges.

[Half-Shot]

N.B. I think matrix_request_seconds is bridge_matrix_request_seconds because the bridge lib prefixes "bridge".

[Half-Shot]

In practise I think we got this wrong. The IRC/Slack bridges use this name to mean requests coming from matrix and the outcome (e.g. did we fail to send a message to IRC, how long did it take).

We found this out when the change got deployed and alerts kept firing because we wern't doing this enough.

I'd propose that we move this Histogram to the event handling portion of the library and report the ourtcome of a processed AS event from a transaction.

[jaller94]

The only incorrect metric is bridge_matrix_request_seconds, right?

[Half-Shot]

Yeah, the others look fine!

[jaller94]

I removed bridge_matrix_request_seconds from this PR. I may come in another PR.