We'd like to look into merging field caps responses on each node, to prevent the coordinator from doing all merging. This will help reduce latency and memory when the field caps request includes many shards.

As part of this work, we can see if we could detect when two indices share the same mappings, and in that case skip over the shards with mappings we've already collected. This can help in the case of time-based indices, where most indices may have the same mappings.

This is a spin-off from #76509 (comment) to capture unresolved work.