copy skeleton only for local filesystems

When the migrator renames a table, for example for upgrade simulation,
it should not rename the primary key to avoid messing up with the diff
because the MySQL Doctrine code expects that index to always be called
"primary".

This reverts commit 73062040.

those apps will have to use the public api

add a function getUserGroupIds for retrieving group ids instead of group objects. this significantly improves performance when using many (nested) groups.

Changes a function call in getUserGroups to only retrieve group ids instead of objects.
this change significantly improves performance when using owncloud with many groups, e.g. nested ldap hierarchy (1.2.840.113556.1.4.1941), since getUserGroups gets called in oc_share::getItems, which is needed for every page request.
in my particular case, it took more than 10s to load the calendar page and more than 6s to load the file page.
this was in an environment with 100 user groups (nested) per user. The performance was bad due to the following call stack:
self::getManager()->getUserGroups($user)
  - getGroupObject() (executed for every group!)
     - groupExists() (resulting in many ldap-requests)
since the groups are loaded from ldap, it is unnecessary to check whether the group exists or not.

The updater is using "before update" repair steps and "regular" repair
steps.
The "regular" repair steps are also used by the CLI tool.

Currently no steps exist but can be added later in the static methods in
the \OC\Repair class.

Added unit test to test messaging, error and exception cases.

This also makes it possible to unit test each repair step class
individually

This reverts commit 482eded8, reversing
changes made to 354cace1.

Changes a function call in getUserGroups to only retrieve group ids instead of objects.
this change significantly improves performance when using owncloud with many groups, e.g. nested ldap hierarchy (1.2.840.113556.1.4.1941), since getUserGroups gets called in oc_share::getItems, which is needed for every page request.
in my particular case, it took more than 10s to load the calendar page and more than 6s to load the file page.
this was in an environment with 100 user groups (nested) per user. The performance was bad due to the following call stack:
self::getManager()->getUserGroups($user)
  - getGroupObject() (executed for every group!)
     - groupExists() (resulting in many ldap-requests)
since the groups are loaded from ldap, it is unnecessary to check whether the group exists or not.

add a function getUserGroupIds for retrieving group ids instead of group objects. this significantly improves performance when using many (nested) groups.

I use the term socket for any extension, either unix socket, or internet socket (port).

I check if the socket is all digits
* only integers 0 and larger would pass this test.
I then check if the string is less than or equal to the maximum port number.

By using "if($socket)" I make sure socket isn't false, empty, or the string '0'.

I don't believe I need to initialize $port because $port will always be set if $socket is true. Please show me if I am wrong here. Thanks

Sometimes MySQL decides to return the shares in the wrong order, but
some parts of the code seem to require the order to be known, at least
so that the parent shares come before the children shares.

This fix adds an ORDER BY clause to force the order by id.

movie previews: use file directly when it's stored locally and encryption is not enabled, fixes #7756