The file used for download in get_product_by_id needs to be re-seeked to start in case of failure. This hopefully closes #16 for true.

Even, modified for coherence with changes in collection names due to use of XML parsed descriptions.

Note that some Perl deps are still missing in main, so an installation
via CPAN is still required even in 12+.

Note that the is_joinable() call check has been also removed, because it would
not block the main program until all threads finished their work and
would cause a premature end of the program.

Now the retry and sleep intervals can be defined elsewhere, just to
avoid having them embedded in the code.

THe former op is used to refresh many information in the stored entries,
it is mainly useful to update hashes of the entries starting from names.
The latter op is used to download missing entries on disk, something
that could for sure happen continuosly due to limitations of the
provided CDSE services.

This is used in refreshing operation in order to update hash values for
existing records.

Note that multi-threaded processing does not like undef fields, so now
OData/ORest and OSearch protocols are dealtw with to avoid emties fields
in resulting hashes.

This is an elementary function to get a full list of the database for
use in refresh/complete operations.

Attributes in REST protos would need a much better management, that is
basically scanning the XML descriptions of collections and get field
names from there instead of using a long and error prone of
manually-defined keywords. There is an open issue for that.