ffmpeg: Maintain higher precision pts between video filter and fps code

ffmpeg: Maintain higher precision pts between video filter and fps code

This is a bit ugly as it attempts to keep most of the computation
in integers before the double based fps code. The use of integers
is to reduce the chances of rounding differences between platforms

Previously the timestamp was rounded to the encoder timebase
before being converted back to double precision which could cause loss
of precision

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>