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.

we should not crash if we try to delete an inexisting document

(closes #17071989)

let serializer decide if cw_etype and doc_type is different