The regression was found in SELECT - OR scenario, which was about 7%-8%
according to QA.

After testing and verification, two changes would help improve the
performance, and current regression is only about 1.7% in the SELECT - OR
scenario, which should be acceptable, because overhead for checking
instant columns are inevitable.

The two changes are mainly about:
1. Two different rec_get_nth_field for record with
(rec_get_nth_field_instant) or without(rec_get_nth_field as is)
instant columns. Thus for the record without instant columns, there won't
be extra checking.
2. Introduce a bit flag in dict_index_t to say if this is clustered index
with some instant columns, rather than consulting with the table.

From the testing, 1 saved the regression a lot, but without 2, it's not
good enough. So it would be nice to pick both. Furthermore, with these
two changes, there are also some small improvement in other testing
scenarios.

RB: 19392
Reviewed-by: Jimmy Yang <jimmy.yang@oracle.com>