The classes ODatabaseDocument, ODatabaseModelImpl and
ODatabaseSource share a single mutex declared as
ModelDependentComponent::m_aMutex.

Commit 403eefe8 introduced
a new deadlock here: in case Yield is called, such as
when the user doesn't have a JRE and a dialog requesting
to install one is displayed, the SolarMutex is released
but the ModelDependentComponent mutex is still locked;
now another thread (such as the AsyncEventNotifier)
can lock the SolarMutex but not the other mutex,
while the main thread can't lock the SolarMutex again.

So the SolarMutex must be on the one hand be
locked before the other mutex, to ensure the behaviour
is the same whether the thread already holds it or not,
and locked after the other mutex, to prevent the Yield deadlock.

One option would be to revisit fca62934
and add back the SfxLibraryContainer mutex, but the code
there is a mess, naively assuming that locking the mutex on
UNO entry points is sufficient, taking no care to temporarily
release it while calling other UNO components or event listeners;
since there's about 5kLOC in all the related classes it
would take some time to rewrite all that.

An easier fix is to remove the ModelDependentComponent::m_aMutex.

This patch removes all locking of that m_aMutex, but it still
exists as a non-shared Mutex because it's needed for the
WeakComponentImplHelper base classes, which shouldn't cause issues.

Reviewed-on: https://gerrit.libreoffice.org/45914Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Michael Stahl <mstahl@redhat.com>
(cherry picked from commit a36b3f37)

Change-Id: I94aaa0ae8698188c98633c638100a309b20976cc
Reviewed-on: https://gerrit.libreoffice.org/45925Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Michael Stahl <mstahl@redhat.com>