[core][metri-agg/2] recommended for all (mostly) data type

[can-anyscale]

This series of PR addresses #47289. The context is that there are metrics such as ray_tasks and ray_actors produce a high volume of time series on prometheus. Our inspection indicates that this is because the high cardinality of the WorkerId field in these metrics in a high scale cluster. See https://docs.google.com/document/d/1AZVZQGGroSbV1w4KG4Vncc0L_tVdpfkM3ueGo_fD9-M/edit?tab=t.0 for the proposed solution.

This PR:

• Small step to complete the RECOMMENDED aggregation level, by further support all metric types except for distributed data. Note that aggregation of distributed data at the node level from worker level are not well defined (or very complicated to be well defined), and the existing set of distributed data are already low cardinality, so I don't spend too much time to support this type of data here. Can be adjusted.

Test:

• CI