The PMIx_Value unload function follows the format of the data
in the pmix_value_t structure. If the data type in the pmix_value_t:

1. is a simple data field (e.g., a size_t), then we just copy the
   data across. The caller needs to provide the storage for the data.

2. is a pointer (e.g., the pmix_envar_t field), then we malloc the
   storage for the value, copy the fields into the new storage, and
   return that storage to the caller. The caller needs to provide
   storage for the pointer to the returned value.

So in other words, the caller needs to provide storage (and a pointer
to) the place where the final data is to be put. If it's a simple data
field, the value will be placed in the provided storage. If it's a
complex data field (e.g., a struct), then we will place the pointer to
that malloc'd data in the provided storage.

The byte_object and regex fields are complex data fields, but were
violating the above rules. So fix those to be consistent, and add
a more thorough check to catch violations.

Signed-off-by: Ralph Castain <rhc@pmix.org>
(cherry picked from commit 0ca2d6c)