Avoids crashing if a rescan is scheduled but the policy module is
destroyed in the meantime

* Every client has a priority based on its role
* For playback, we allow only a single client to play at a time
* For capture, we allow all clients to capture simultaneously
* Every time the "selected" device changes (either because devices
  are discovered/removed or because the user changed the selection),
  the clients are re-linked to the new "selected" device.
* When a playback client quits and there are others waiting unlinked,
  the highest priority one is linked automatically.
* This also properly fixes re-linking the correct client(s) to the
  correct device(s) when wireplumber exits and restarts.

This allows linking the client to the stream that has the same name
as the role that it requests.

It so happens that sometimes the client proxy is created and destroyed
immediately, almost instantly, which causes this code to crash in case
the proxy pointer is gone when our idle callback tries to destroy it.

This change makes the whole operation safe.

Currently, the stream names have to be duplicated in the configuration
file. This is going to change in a future version.

... and simplify some code there

This is currently the documented way of describing controls
that have an effect on all streams

This is a bit of a hack, temporarily needed for AGL.

Currently this will just grant full access to all clients.
A future version will refine that to grant access to specific
objects only.