Benchmark the field_caps API #84504
Description
We have recently improved the field_caps API execution in the effort of guaranteeing that its performance is reliable. See #79212 and #83494 among others. We have seen that field_caps can be a bottleneck when it hits many indices and many fields, especially as it is heavily used by Kibana which relies on it even before executing any search. Part of this discussion is captured in #83494 .
We should complete this workstream by working on benchmarks that measure the performance of field_caps, executed locally as well as against remote clusters. The goal is to catch future regressions as well as make recommendations around how well field_caps can scale, e.g. what level of performance can we guarantee with a high number of indices (hundreds?) and fields (thousands?) ?