Use an unique stored field name for each ignored source entry

[martijnvg]

Today when loading ignored source for a specific unmapped fields, then all ignored source entries are processed/parsed. This has a significant impact on read performance of ignored source. One reason that this works this way is that al entries are stored using the same stored field name (_ignored_source).

If each ignored source entry has stored field entry with its own name (e.g using _ignored_source_* as prefix) then we can better control which entries need to be processed / parsed. For example if only one ignored source entry is requested, then we don't to process/parse all entries to figure this out.

Before this wasn't much of an issue given that most fields would be mapped and therefor not end up in ignored source and typically when synthesizing source all ignored entries need to be processed/parsed.

This related to #130886 and may depend on this.

[elasticsearchmachine]

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

[jordan-powers]

Opened a draft PR, still a WIP though: #132142

[jordan-powers]

#132142 is merged, this feature is now implemented behind a feature flag. I'll leave this issue open until the feature flag is removed.

[jordan-powers]

This was implemented in #132142. However, after looking at some benchmarks and discussing with Alan and Martijn, we realized that splitting ignored_source like this creates too much overhead (due to fieldinfo merges) when there are a large number of unique unmapped fields. So this was reverted in #133839 and won't be implemented.