Investigate using specialized field visitor when loading ignored source.

[martijnvg]

Today when reading stored fields in compute engine the https://github.com/elastic/elasticsearch/blob/31c2594bb96d47748b29ee923706312b01fd547e/server/src/main/java/org/elasticsearch/index/fieldvisitor/StoredFieldLoader.java#L208 gets used via:

elasticsearch/server/src/main/java/org/elasticsearch/index/fieldvisitor/StoredFieldLoader.java

Line 208 in 31c2594

this.visitor = new CustomFieldsVisitor(fields, loadSource);

This has a number of inefficiencies when loading ignored source for all documents that match the query. Like interacting with hash map / hash set / parsing id when a document is processed. This is ok in the context of the fetch phase in search api, however in the context of loading ignored source in esql not.

Looks like ~10% can be improved here.

[elasticsearchmachine]

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

[elasticsearchmachine]

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