XContentMapValues.extractRawValues() contains a hack whereby we
ignore extra 'dummy' dotted path values appended to a source path - ie,
loading data from foo.bar.dummy will also load from foo.bar. This was
used to implement source loading for multifields, which do not have a name
that corresponds to an existing path in source data. This was safe in general
because you can't have a field that is a dot prefix of another field in index
mappings.
With runtime fields, however, this changes. You can define scriptless runtime
fields for both foo.bar and foo.bar.dummy, and it is important that the
latter does not pick up values from the former path.
Source loading for multifield highlighting has already moved away from using
this hack by loading everything through value fetchers. This commit removes
the hack entirely, and updates term vector loading to use a value fetcher
as well
8806c1a6d9