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

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

closes #17079908

* 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.

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

(closes #17071989)

same thing for --except_etype

Also : ccplugin - generator serves all actions, not split up by type
       es.py - add CUSTOM_ATTRIBUTES mechanism