TokenStream created from a term vector field. The term vector requires positions and/or offsets (either). If you
want payloads add PayloadAttributeImpl (as you would normally) but don't assume the attribute is already added just
because you know the term vector has payloads, since the first call to incrementToken() will observe if you asked
for them and if not then won't get them. This TokenStream supports an efficient
reset()
, so there's
no need to wrap with a caching impl.
The implementation will create an array of tokens indexed by token position. As long as there aren't massive jumps
in positions, this is fine. And it assumes there aren't large numbers of tokens at the same position, since it adds
them to a linked-list per position in O(N^2) complexity. When there aren't positions in the term vector, it divides
the startOffset by 8 to use as a temporary substitute. In that case, tokens with the same startOffset will occupy
the same final position; otherwise tokens become adjacent.