NvEnc Limit Stop Testing for device -- Various Fixes

[nathangur]

Issue relating to this PR: #27

• Better Nvidia driver limit checks with a now persistent known limit of the GPU driver across tests.

• Fixed some weird cuda stuff

• Use ffprobe now for checking the files

• More comprehensive and in-depth ffmpeg error logging

• Fixed MaxStreams for Nvidia GPU's always being 1 after a test was finished

Possible issues:

On my machine the CPU likes to jump from 1 worker to 9 workers, no clue if it's just really excited to be benchmarking or if it's just bugged.

[BotBlake]

Note: The high jump in workers is expected.
This is part of our "worker upscaling" logic.

For upscaling the amount of workers, we generaly take a look at the speed of the run.
So e.g. :

Run: 1, Workers: 1, Speed: 5.2x
-> Workers = Workers + Speed
Run: 2, Workers: 6, Speed: 1.3x
[...]

We also have downscaling measures that will compensate for too heavy upscaling.

I assume this is the reason that your GPU jumps from 1 workers straight to 9.
Same should also happen on GPU tests, if the GPU is powerfull enough.

That way we save loads of time compared to upscaling by 1 every run

[BotBlake]

Note:
The Script will Scale Up the Worker amount, until some sort of failure happened.

In some specific conditions we Scale down to verify that we found the absolute Maximum of streams.

Adding a value to failure_reason implies that a failure has happened!