This patch needs more testing, I think.

Some instruction sets can't divide (e.g. ARM), but pretty much all of
them can shift bits.

dts spdif output was broken by the switch over to the new filter
API.

  * dtstofloat32 should only take control if there is _no_ spdif
    output requested otherwise the decoded stream ends up on the
    spdif device
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

A/52 spdif output was broken by the switch over to the new filter API.

  * a52tofloat32 should only take control if there is _no_ spdif
    output requested otherwise the decoded stream ends up on the spdif
    device

  * a52tospdif has to set a pts/dts value otherwise the frames get
    (wrongfully) discarded as too late because a wrong pts/dts value
    got evaluated by aout_OutputNextBuffer()
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

It just allows disabling their use.

 They now all mean that inline assembly (asm volatile (:::) for a specific
instruction set can be used.

(This is a really tiny fix but I'm used to use compile script).

This reverts commit ea4b6069.

config.h should be bundled with dev. packages (this concerns at least
libvlccore-dev on Debian). Or regenerated by a script.

"psz_af2" -> "psz_af"

This follows the historical LibVLC semantic (and new --rate one), so
bigger is faster. In the process, simplify a few code paths, especially
those in LibVLC, but make many other code paths more complicated due to
integer<->float conversion. Most of those paths could probably be
simplified.

 * Fix RC "fastforward" command behaviour (it was slowing down instead)
 * Fix str_format_mera 'R' format string with negative rates and
   rates such that (rate % 1000 < 100).

python-ctypes: implement shortcut media.add_options (in addition to the existing instance.media_new( mrl, options...))