1- We can cache the result of the first xbps_pkgdb_init() when it fails
and avoid the malloc/free/access from it.
2- We cache the uname(2) result into a private var in xbps_handle and
use it in xbps_pkg_arch_match().
This improves performance by ~5% approx and it's close as it was before
introducing the repository index format 1.5.
This fixes xbps_match_pkgpattern_in_array() which had the args swapped
in xbps_pkgpattern_match() resulting in inverted results while resolving
virtual packages in a transaction.
