Cut 1.20.5 with the revert of #1424 (compareMetricFamilies)

[bwplotka]

I spoke to a few people (e.g. @dashpole @dgrisonnet @kakkoyun @ArthurSens @vesari @logicalhan), and it feels it might be easier for everyone to revert.

cc @machine424

Rationales:

• We feel the pain of the subsequent changes in Kubernetes and other places (e.g. Google). We want to make sure downstream users trust client_golang as a stable and minimal dependency.
• Reverting this change should NOT break (re-break) users who fixed their tests and switched to a new behaviour.
• We broke users who used this API to test for a metric to be absent without replacement (thanks @dgrisonnet for innovating here in a PR).
• Based on @dashpole work on the upgrade, this change didn't provide any major "improvement" in Kubernetes tests so far.

I propose we revert this, and work on a long term solution. With the team we also discussed a way, where every client_golang release pipeline could propose an upgrade PR to a Kubernetes library that heavily depends on client_golang and check if the CI passes (compatibility check with a big user) -- to avoid similar issues in the future.

[ArthurSens]

Reverting this change should NOT break (re-break) users who fixed their tests and switched to a new behaviour.

LGTM as long as the above was properly tested 😅!

[ArthurSens]

Impressive that we've got a fifth patch release this time. As discussed in our call today, creating release candidates would improve everybody's life by a ton!

Issue opened here: #1646

[bwplotka]

LGTM as long as the above was properly tested 😅!

Yea, we only remove a check, so except this being more error-prone (we will reopen #1351), those who already did a work to adjust their test should be not impacted. However, for empirical test purposes, I kindly asked @krajorama to create a quick PR on Mimir with PR to double check (:

EDIT: One test failed grafana/mimir#9624 (comment), but only because a different error message was produced. I think it's acceptable or not as smooth as I expected. Seems like still it's more annoying to keep this change vs revert, so proceeding.