Rescue from network errors when querying Prometheus in the background

What does this MR do and why?

Currently PrometheusAdapter can fail to update its reactive cache when encountering unhandled network errors.

This is because only a subset of all possible errors in Gitlab::HTTP::HTTP_ERRORS is handled by Gitlab::PrometheusClient#get.

This MR rectifies this by making PrometheusClient reraise all HTTP_ERRORS as PrometheusClient::Error.

Screenshots or screen recordings

Refer to: #341633 (comment 714590480)

How to set up and validate locally

Refer to: #341633 (comment 714590480)

MR acceptance checklist

This checklist encourages us to confirm any changes have been analyzed to reduce risks in quality, performance, reliability, security, and maintainability.

• I have evaluated the MR acceptance checklist for this MR.

Related to #341633 (closed)