re: Github issue Unidata#1956

The function NC_compare_nc_types in libdispatch/dcopy.c uses an
incorrect algorithm to search for types. The core of this is the
function NC_rec_find_nc_type in libdispatch/dcopy.c. Currently
it searchs the current group and its subtree.

Additionally, the function NC4_inq_typeid in libsrc4/nc4internal.c
has been extended to handle fully qualified names. It was originally
designed to do this, but for some reason never completed.

The NC_rec_find_nc_type algorithm has been altered to match the
algorithm used by NC4_inq_typeid. It operates as follows.

Given a file F, group G and a type T. It searches file F2, group
G2, for another type T2 that is equivalent to T.

The search order is as follows.
1. Search G2 for a type T2 equivalent to T.
2. Search upwards in the ancestor groups of G2 for a type T2 equivalent to T.
3. Search the complete group tree of F2 in pre-order, breadth-first order to locate T2 equivalent to T.

Also add a test case to validate algorithm: ncdump/test_scope.sh.

Note, this change may cause compatibility problems, though it is
unlikely because two different equivalent type declarations in
one dataset is unlikely.