Now audio codec frame size will not affect latency precision, at the
small cost of one memcpy per frame.
With the maximum number of OpenSLES buffers (255), we can have a precision
of 4ms with a buffer size of ~1 second.

We can access it with GetState

These objects do not depend on the audio format

To fix type aliasing, we need to know whether the data is float or integer,
or use only char pointers.

Also add support for FL64 and arbitrary lengths.

The FLAC decoder aligns samples to the right. So shift the samples to
to the left.

Ref #2744 #4252
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

minimal_macosx: remove dependency on the Carbon framework by replacing SetSystemUIMode calls by their Cocoa counterparts

remnants to be removed in next commit

Endianess is now converted by the raw decoder/encoder.

It seems S24N needs to be kept for the FLAC decoder though.