audiosink: doesn't wait for ringbuffer to drain on EOS with sync=false

Submitted by Topolsky

Link to original bug (#764120)

Description

Created attachment 324653
concat cuttoff last part of second source

Concat element somehow cuts off 2 seconds of end of second input.
This can be triggered out by analysing output of this pipeline:

gst-launch-1.0 concat name=c audiotestsrc num-buffers=200 ! queue ! c. audiotestsrc num-buffers=200 volume=0.5 ! queue ! c. c. ! alsasink buffer-time=1000000 sync=false

The bug is easily reproducible and it happens every time.
I use branch 1.6 from 15.3 (gstreamer - commit c5ad081b9 about baseparse)

I recorded output of soundcard with audacity and I see that last part of second source (with volume 0.5) is much shorter than first source. Both audiotestsrc has num-buffers=200 so they should be equal.

Attachment 324653, "concat cuttoff last part of second source":