So we don't contend for the global structure lock all the time.

Introduced in the previous commit, pointed by Pierre. Thanks!

If it happens, there is a bug, and we'll assert in _yield.

Besides, we assert that this isn't the case right before...

When i_refcount drops to zero (under the structure lock), the object is
removed from the objects table before the structure lock is released.
Hence, no objects with a nul refcount can ever be left in the table.

There is an unlikely race condition if another object is removed from
the table right between we unlock the structure lock but before we
return from vlc_object_get() (so that the i_middle offset gets wrong).

Plugins are already using pl_Yield to get the playlist, as they should.

(Could someone please review these commit)

Now we can remove the VLM object type.

Afterall it's not our bug. More importantly, the vlc_cond_wait "possible
deadlock" debug message has been removed, which makes the problem a lot
less likely to happen.

In RTSP/1.0, don't return any range (as far as I understand).
In RTSP/2.0, we'd have to compute the time from the beginning.

(was not in previous release, so not adding _obsolete).

The bug is present in original "Linux DeCSA Tool" on http://csa.irde.to/