Context:
This is a follow-up to OPT-2831 where the current implementation of the grouping was optimized by removing redundant calls.
However the optimal implementation would actually be based on doing this aggregation in elasticsearch and not having to load all definitions into optimize to then do this aggregation manually.

AT:

• definition grouped by version and tenant aggregation is implemented using an elasticsearch aggregation, ideally in a shared way in the new generic DefinitionReader
• performance test is changed so that 50 definitions (5 versions) and 5 tenants ensuring a response time of <1s

Note:
With OPT-2849 a generic definition reader that uses aggregations was already introduced, it doesn't have support for version information yet though.