... in case the global is removed from the registry before
the initial augment completes

When a new global is created, it is not certain
if the registry global event or the proxy bound event will
be fired first. In order to make sure we associate all
proxies to their WpGlobals correctly, we now wait a core sync
before exposing globals to the object managers, so that in case
the implementation proxy receives the bound event after the
registry creates the WpGlobal, we can make sure to use this
proxy instead of constructing a new one through the object managers

wp_exported_export() used to do that internally, but it's cleaner
to do it this way now

There are 3 kinds of WpProxy objects:
 * the ones that are created as a result of binding a global
   from the registry
 * the ones that are created as a result of calling into a remote
   factory (wp_node_new_from_factory, etc...)
 * the ones that are a local implementation of an object
   (WpImplNode, etc...) and are exported

Previously the object manager was only able to track the first kind.
With these changes we can now also have globals associated with
WpProxies that were created earlier (and caused the creation of the global).
This saves some resources and reduces round-trips (in case client
code wants to change properties of an object that is locally
implemented, it shouldn't need to do a round-trip through the server)

* core no longer exposes create_remote/local_object
* node, device & link have constructor methods
  to enable the create_remote_object functionality
* added WpImplNode to wrap pw_impl_node and allow creating
  "local" node instances
* added WpSpaDevice to wrap spa_device and allow creating
  "local" device instances
* exporting objects in all cases now happens by requesting
  FEATURE_BOUND from the proxy, eliminating the need for WpExported
* replaced WpMonitor by new, simpler code directly in module-monitor
* the proxy type lookup table in WpProxy is gone, we now
  use a field on the class structure of every WpProxy subclass
  and iterate through all the class structures instead; this is
  more flexible and extensible

+ use the pw_proxy API to find the bound id instead
of relying on WpGlobal

This has the advantage that it works also for exported
objects and for objects that have been created by calling
into a remote factory (such as the link-factory), so we can
now know the global id of all proxies, not only the ones
that have been created by the registry.

This is not used anymore. It was useful when we were using it as
a detail in the global-added signal, but that is gone now.

They are equivalent, there is no real benefit in having both

This module allows wireplumber to create static nodes that match a specific
device using a spa node factory. Matching is optional, and if there is no match,
the node will always be created.

This ensures that endpoints with keep=false will still have
a chance to link when ones with keep=true exist. This also effectively
means that role priority does not matter when keep=true: we keep these
links active no matter what, while policy still applies normally for
all others.

For example, a final sorted list with these endpoints will end up
looking like this:

* ep1, role priority=25, keep=false
* ep2, role priority=20, keep=false
* ep3, role priority=25, keep=true
* ep4, role priority=75, keep=true

... which will effectively cause ep1, ep3 and ep4 to be linked.

keep=true should work in both ways:
* keep the endpoint with this property linked at all times
* keep other already linked endpoints when linking this one

This is no longer needed since we sort endpoints by role priority
before trying to link them and we link only the highest priority one.
After this sorting, the highest priority endpoint is guaranteed
to be able to link, so _can_link_stream() always returns TRUE.