because otherwise the metadata unit test fails

* Implement the proxy properly with info caching, methods, etc
* Get rid of useless variables and includes

meson.build:
When the 'wrap_mode' option is set to 'nodownload' use a system version
of cpptoml. This does not require using git and having a network
connection during build, which is important for Linux packaging
infrastructure.

subprojects/cpptoml.wrap:
Pin revision to last release tag (v0.1.1).

lib/wptoml/*.cpp:
Remove 'include/' prefix from all cpptoml related includes, at is not
required.

Closes #17

Related to #18

* add library.name to not require adding `add-spa-lib` in pipewire.conf
  or wireplumber.conf for this to work
* add a commented local=true; it can be useful to run those nodes
  locally for testing, sometimes

there are underlying issues with the state management of the graph,
so it's not a very good idea to rely on it to activate / deactivate
our convert node depending on the links that exist

instead, track the links ourselves and create/destroy the corresponding
links to the target node accordingly; it's a more robust approach

it was also not entirely correct that we would previously configure
the convert node to be a driver (node.driver=true); maybe this caused
the underlying issues in the first place... we don't need it now
anyway, so it's gone

Just like media-session does
JACK nodes should be left untouched

1. device export proxies must be destroyed manually since they are
   not associated with the WpRegistry
2. the monitors should not disconnect before all WpSpaDevice objects
   are destroyed; remove the manual disconnect call and let GObject
   ref counting do its job (the core will disconnect when its last ref
   count is dropped after the last monitor plugin is destroyed)

* do not copy the full alsa node properties set
* use a node description that makes the nodes look better in JACK
* use . instead of / as a separator for the node.name, like elsewhere
* add audio.convert spa lib association in the tests that use si-convert;
  previously it used to work because library.name was present in the
  properties copied from the adapter (and it so happens that the adapter
  lives in the audioconvert spa plugin as well ...)

JACK uses : as a separator to distinguish the port name
from the node name, so it ends up doing wrong separations
if we have : in the node name

if there was a remainder in the buffer from a previous iteration,
make sure not to fill too much data and go past the end of the buffer

after consuming the whole buffer, cur & linestart need to
be reset to point back to the beginning of the local buffer

This may happen if the plugins were loaded but not activated
due to some parse error in wireplumber.conf

previously, we would only remove the endpoint associated with
the node, but not the monitor endpoint...

to implement that, we now store the node's associated session items
in the node's qdata instead of a local hash table, so that we
can store an arbitrary ammount of session items per node

otherwise the parent never un-exports or de-activates itself
and we leak a bunch of stuff

there is no need to add self in the transition's data,
it's already in source_object

This fixes a failure to activate the monitors if we are using
si-adapter directly instead of si-audio-softdsp-endpoint

Requires PipeWire < 0.3.3 or > 0.3.6 to actually work