Avoid to create a new PA context for each new client by using a hash
table containing the list of ring-buffers and the shared PA context
for each client. Doing this will improve application memory usage in
the cases where multiple pipelines involving multiple pulsesink
elements are used.

Fixes bug #624338.