* Avoid creation of map with all archive contents; instead parse packages
while walking though the archive
* Avoid instantiation of std::string in come cases (using std::string_view)
* Reuse libarchive's archive entry when walking though archive
* Use visitor-based database parser in all places to avoid intermediate big
array with all package objects
Especially when enabling files DBs it is quite problematic to store all
package objects of a database within one big array. This change avoids the
array and instead adds the packages directly. The disadvantage is that
clearing the database isn't as simple anymore.
* Only lock the config for writing the reloading the config file
* Make sure all write operations to the database acquire an "update mutex"
to ensure only one write operation happens at a time
* Do *not* acquire any additional locks when reading from a database as it
should be safe to do so even when a write operation happens because
* LMDB read and write transactions can happen at the same time
* The package cache has its own mutex anyways
* Write ops to the package cache try to lock the "update mutex" to
prevent writing "old" data to the cache during updates
* Make "lastUpdate" atomic to avoid locking the config when accessing it
* Do HTTP head request first when loading database from mirror to avoid
downloading the full database all the time
* Use the last modification date of the local database file because with
the persistent storage even local database reloads became a bit expensive
When `repo-add` fails but files could be copied before refusing to override
is actually very annoying because one has to delete the files before
restarting the build task.
* Record full path for the deletion of orphaned signatures
* Test for the existence of the signature symlink itself and not the target
because the symlink target (in the any directory) might not exist anymore
but the symlink should be removed in any case
