Newer GnuTLS versions do it internally and in a more correct fashion.

Certificate date verification has changed. To get proper errors,
we need to depend on the newer versions.

Fix #5433

These variables are overwritten a few lines later.

(for whatever reason this was only observable on snow leopard for me)

DecoderWaitDate() releases the lock and thus state can change.
In particular, the aout can be destroyed, or paused or resumed.
So those checks must be done after DecoderWaitDate().

There may be a cleaner way to do this, but it will likely be more
invasive. This patch fixes #5825 and probably #6369 too.

A discarded buffer is not necessarily in the future. It might simply be
that there is no output or that the playback rate is out of range.

DecoderWaitDate() needs to return with the lock held to fix #5825.
Anyway, there was not much point in releasing the lock after
DecoderWaitUnblock() and acquire it again without sleeping.

aout_DecPlay() needs the lock, since the aout is not reentrant.

Therefore, add a copy to extras to prepare the VLCKit split

Depends on --enable-small and HAVE_NEON
If we use NEON, we assume at least cortex-a8 CPU which means thumb2 is supported

Authors include:

Pierre d'Herbemont
Felix Paul Kühne
Faustino Osuna
Rafaël Carré
Sebastien Zwickert
Rune Botten
Toralf Niebuhr
Jean-Baptiste Kempf
Jean-Paul Saman
Mike Schrag
Pavlov Konstantin
Tanguy Krotoff
Andrey Utkin
Daniel Mierswa
Derk-Jan Hartman
Malte Tancred
Pere Orga

Due to his passing into the next world, Emmanuel de Roux's code was
not relicensed and has been removed.
May you remember him and have a thought for his family.

The destructor called interaction_Unregister(), thus also call interaction_Register()

Reported by: An anonymous contributor working with the SecuriTeam Secure
Disclosure
program (http://www.beyondsecurity.com/ssd.html)

The PICTURE_RING_BUFFER used a fixed size, which when running out of available pictures
produces visible artefacts in the encoded stream. Allocating a new picture and keeping
track of them in a fifo is much simpler (in the multiple threads encoding scenario) and
solves the ring buffer overrun.

 * Use assertions for argument checks
 * Add missing error checks