HLS can have inconsistent numbering through playlists.
In that case, a datetime is used to provide date and time reference
for the first segment.
We need then to convert our segment numbers using that
reference to be able to switch to the correct segment
number.

Exemple:
http://b028.wpc.azureedge.net/80B028/Samples/0e8848ca-1db7-41a3-8867-fe911144c045/d34d8807-5597-47a1-8408-52ec5fc99027.ism/Manifest(format=m3u8-aapl-v3)

Some streams have microsecond durations, > default timescale

Ref #14885 #14166 #14162 #14442

It makes it more clear that text_layout* and platform_fonts* are
Freetype2 specific

freetype: drop CoreText specific Select in favor of the generic implementation and add family lookup

No fallback yet

This removes a Carbon dependency and adds support for tvOS and iOS

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

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

Drops the current long term average to a moving
average/vertical horizontal filter with an
observation window of 250ms.

Decoder still need to be signaled when FIFO is already empty.

This reverts commit 32040fdc.

Since recent changes, DecoderThread need to be signaled only when
frames_countdown is incremented.

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

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