This new version of the patch provides a more general solution for the unknown field NPE problem (the solution for the delete problem is the same and is included in the new patch file).
For the NPE problem, the new patch handles the case of an unknown field explicitly in all ParallelReader API's for which it is relevant. This is quite a few of them, but I think I got them all.
The first solution of specifying the field list for each reader was expedient in my applicaiton since all fields are pre-configured and validated on index operations. However, for Lucene in general such field processing is not required. E.g., with ParallelReader as it is in Lucene now and with the first version of the patch, a user-entered misspelled field name passed in a Query to search() generates an NPE instead of zero results. With the new version of the patch, zero resutls are returned properly.