Add bwc layer for 'romanian' analyzer

The 'romanian' language analyzer has been improved in Lucene 10 in two important
ways. First, the snowball stemmer has been modified to work with s-comma and t-comma characters
but only with their cedilla forms used when Romanian didn't have full Unicode
support (snowballstem/snowball#177). Second, the
analyzer now contains a normalization step to map cedilla forms to forms with comma.

In order to maintain backwards compatibility with existing indices, this change
moves the Lucene 9 stemmer over to the analysis module was a deprecated variant
and creates the analyzer for existing indices with the "old" stemmer and without
the normalization step. New indices automatically run with the improved
behaviour.