Follow-up from "Add DB metric for external audit streams"

The following discussion from !184339 (merged) should be addressed:

• @praba.m7n started a discussion: (+1 comment)

  thought: I really think GitlabServicePingWorker can have delayed data_consistency instead of always (which targets the primary).

  Reading ~14GiB of buffers is a lot, I am not sure about the timing guidelines for ServicePing as this is once in a week event. But in general it's good to not to run this on primary.

  Can be considered non-blocking/out-of-scope if we manage to get the query to run faster.