ESQL: Group by all optimization

[leontyevdv]

Normalizes ordinals so that each unique tsid uses the first ordinal encountered for that tsid.

The FirstDocIdGroupingAggregatorFunction collects the first doc id for each group. With time-buckets, the same tsid can appear in multiple groups. When loading the dimension field, this may result in loading multiple documents for the same tsid several times.

There are two options:

1. Load only one document per tsid and apply tsid ordinals to the dimension values. This requires a separate value source reader for dimension fields, or modifying the current reader to understand ordinals and apply them to the dictionary.
2. Remap the group ids in the selected vector so that all groups with the same tsid use a single group id. This means we
   may load the same document multiple times for the same tsid, but not different documents. The overhead of loading the
   same document multiple times is small compared to loading different documents for the same tsid.

This implementation uses the second option as it is a more contained change.

Example:

_tsid key:     [t1, t2, t1, t3, t2]
selected:      [0,  1,  2,  3,  4]
first doc ids: [10, 20, 30, 40, 50]

re-mapped selected:            [0, 1, 0, 3, 1]
first doc ids with re-mapped : [10, 20, 10, 40, 20]
Loading docs: [10, 10, 20, 20, 40], which is not much more expensive than [10, 20, 40]

Part of #136253

[elasticsearchmachine]

Hi @leontyevdv, I've created a changelog YAML for you.

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[elasticsearchmachine]

Pinging @elastic/es-storage-engine (Team:StorageEngine)