If this is not done, elasticsearch.scan method retrieves the old index values
so wrong values will be copied while synchronizing different indexes.

We probably should do it by default in `indexer.es_index` method.
We probably should do it in `indexer.es_delete` method.

see also https://www.elastic.co/guide/en/elasticsearch/reference/current/docs-refresh.html

related to #17090555

When indexing an entity type, don't generate a single
and possibly huge rset with all entities of that type.
Instead, fetch entities chunk by chunk and free caches
between each round.

closes #17090555

cubicweb_elasticsearch/migration/* was not included in source package.

ESTransactionQueue batches all ES operations on postcommit.

Client cubes could override the default logic to add, log,
ditch… any other operation.

mocked parameter list was inverted

Indexer instances are the ones who know how to connect
to elasticsearch and which index should the query should
be executed on

As for ES "id" parameters, "doc_type" should be customizable,
there is no reason why es document type should match cw_etype

and it seems redundant with `allow_all_hooks_but`

Default is to use 'eid' but one may want to use something
else (stable id, ark, etc.)

closes #17079908

For optimization sake, just make sure the original rset
fetches everyting or use complete=True but relying on
cw_attr_cache will cause very obscure bugs

closes #17079907

* get rid of CUSTOM_ATTRIBUTES. Replace it with IFullTextIndexSerializable adapter `custom_indexable_attributes`
  attribute;
* FullTextIndexSerializable.serialize method : default complete value is now True*


closes #17079905

otherwise the import just hand forever.

First stop removing them (was done for legacy cube) then register them.

since it's a cube as package which are supported from there.

CubicWebConfig.config_for might accept extra parameters such
as "debugmode"

--debug and --dry-run should be used as flag-only options.
They don't expect a "yes" / "no" value.

Also, remove tests on stdout in debug mode because it's
fragile and messes up with logging handling in cubicweb.

Besides from being a debug-only trace, it raises an error when
no actual document is indexed.

It's not used (and therefore not real-life-tested) anywhere
and current implementation is known to be partial and a bit
hard to maintain.