ffprobe uses buffer while checking frames

[WoZ]

So this issue gives incorrect values for the first analysis via method processFrame() (and future processFrames.encoderStats() and processFrames.networkStats()) methods.

[WoZ]

Btw, PR I've added brokes functional tests. That's because previously FramesMonitor received a full set of audio and video frames from the beginning of a mp4 test file (ffmpeg -listen 1 waits for incoming connection to start stream the file). That buffer helped to calculate video and audio frames.

If we add -fflags nobuffer first frame received by monitor may vary from platform to platform and from version to version of ffmpeg/ffprobe.

I see few solutions:

• remove failing functional tests
• replace failing test files with new ones and rely deeply on some frame, let say, at 5 sec from the beginning of file.

The second approach will work only when ffmpeg/ffprobe can determine metadata of the stream prior detection of "control" frame. Additionally, -probesize may help to prevent overflow but I think that this is bad idea to tune -probesize. I saw a lot of segfaults with incorrect values and probesize may cause false negative results.

Guys, what do you think? @fox-hellraiser @OlegPoberegets

[fox-hellraiser]

@WoZ

remove failing functional tests