Av1transcoder

- Fixed crash with newer ffmpeg versions, caused by ffprobe changing it’s XML output format.

- Fixed parser issue with custom ffmpeg arguments that contain nested quotation marks with spaces
- Clarified wording in the --help output.
- Marked as compatible with Python 3.8

- Fixed issue with deleting temporary files at the end of the encoding process
- Write encoder parameters to the matroska container using a custom metadata tag.
The full encoder settings string is written to the VIDEO_ENCODER_SETTINGS tag.
This should make it easier to determine the used encoder parameters in the future.

- Prepare public release
- Improved README, --help output.

- New option --limit-encodes NUMBER. Only run up to NUMBER scene encodes in total. If the limit is hit, no new scene
encodes will start and the program will exit. All scenes in all input files count towards this limit. Merging into
the final output result will not be performed, if this option causes the scene list to be incomplete. For the sake
of this option, the two encodes required for a Two-Pass encode of a scene count as one encode towards this limit.
- New option --crop TOP BOTTOM LEFT RIGHT. Uses the ffmpeg crop filter to trim the given pixels off the video border.
Can be given multiple times. If so, each input file gets it’s own crop parameters in the order both are given on the
command line. If more input files are given, the last --crop parameters are used for all remaining input videos.
- Sanitize user input for numbers. Options requiring positive numbers now check that no negative numbers are supplied.
The Scene cut threshold now checks that the input is in the range (0, 1].
- Updated the --help output
- Updated the default encoder parameters. The default preset is now tuned for high quality SD encodes.

- Split first pass from second pass for two-pass encoding and use a different task scheduler for two-pass encodes.
- Use the first-pass log size as simple metric to estimate
second-pass runtime and schedule the second passes accordingly.
Large logs indicate long and/or complex scenes that take long to encode.
- Start encoding long running scenes first, which will result in better multicore usage at the end of the processing.
- It avoids starting long scenes, like the credits, at the end of the processs, and therefore lessens the impact of
a single, long encode delaying the whole process. With this scheduling approach, it is way more likely that the
last running encodings will be encoding short and easy scenes and therefore having less overall delay.