bufr_subset_find_values() fails when TLC keys are used

[cpb]

Testing with the following template:

  201013 201023 5012 101003 12001

Shows that if I do a find using a TLC code, ala:

         k = 0;
         bufr_set_key_location( &(codes[k++]), 5002, 64 );
         bufr_set_key_int32( &(codes[k++]), 12001, NULL, 0 );
         n = bufr_subset_find_values( dss, codes, k, 0 );

I get the wrong instance of 12001. The bug appears to be that in bufr_subset_find_values(), a TLC match still increments to the NEXT descriptor in the list, and because that descriptor is replicated, I get it instead of the one which had the matching TLC value.

I'll commit a test case...

[cpb (chris-beauregard)]

Ah... another bug is that:

         k = 0;
         ivalues[0] = 291;
         bufr_set_key_int32( &(codes[k++]), 12001, ivalues, 1 );

         n = bufr_subset_find_values( dss, codes, k, 0 );

Gives me the *first* 12001 descriptor in the sequence, because when we match descriptors but not the value, we don't reset the jj index. Which means we don't set it to the actual match.