Add enrich source field mapper.

[martijnvg]

The enrich source field mapper stores the source of a document as
binary doc values. This is useful in cases where retrieval speeds
are more important than compact storage (which is what SourceFieldMapper does),
which is the case for the enrich processor.

Prior to this change enrich processor was using _source stored field
to fetch the enrich document to enrich document being ingested.

The enrich policy runner, when creating the enrich index, disables
_source meta field and enables the _enrich_source meta field.

The enrich source field mapper is an internal field, which is only
meant to be used by the enrich feature.

Relates to #41521 and #32789

[elasticmachine]

Pinging @elastic/es-core-features

[martijnvg]

Running benchmark using this track with this new meta field type already shows good performance improvements:

|                                                Min Throughput | insert-messages-with-enrich-pipeline |    11260.4 |       19572 |  8311.57 | docs/s |
|                                             Median Throughput | insert-messages-with-enrich-pipeline |    15292.7 |       23454 |  8161.33 | docs/s |
|                                                Max Throughput | insert-messages-with-enrich-pipeline |    16502.6 |     24143.1 |  7640.51 | docs/s |
|                                       50th percentile latency | insert-messages-with-enrich-pipeline |    553.256 |      356.66 | -196.596 |     ms |
|                                       90th percentile latency | insert-messages-with-enrich-pipeline |    687.964 |     567.713 |  -120.25 |     ms |
|                                      100th percentile latency | insert-messages-with-enrich-pipeline |    814.368 |     763.496 | -50.8725 |     ms |
|                                  50th percentile service time | insert-messages-with-enrich-pipeline |    553.256 |      356.66 | -196.596 |     ms |
|                                  90th percentile service time | insert-messages-with-enrich-pipeline |    687.964 |     567.713 |  -120.25 |     ms |
|                                 100th percentile service time | insert-messages-with-enrich-pipeline |    814.368 |     763.496 | -50.8725 |     ms |
|                                                    error rate | insert-messages-with-enrich-pipeline |          0 |           0 |        0 |      % |

(the baseline is what is in the enrich branch and the contender what is in this PR)

[jpountz]

It looks good. Something I'm wondering is if we can prevent users from using this field on their regular indices?

[jpountz]

you can use FieldData#singleton(values)

[martijnvg]

Thanks for reviewing @jpountz!

Something I'm wondering is if we can prevent users from using this field on their regular indices?

Perhaps enforcing that the index name should start with .enrich- is enough? All enrich indices will have this prefix (EnrichPolicy#ENRICH_INDEX_NAME_BASE).

[jpountz]

👍

[martijnvg]

This change is no longer relevant.