[WIP]: Deterministic Entropy Coding Cross-Architecture

[jbrough]

Please anyone feel free to help with this :)

tested encoding on OSX and decoding on Linux AMD:

% of N second chunks decoded and playable without errors (Binary search failed, end of stream etc).

encodec vs PR branch

1 second: 90% vs 98% 10 second: 60% vs 90% 30 second: 14% vs 85% 60 second: 0% vs 75%

This is a bit skewed as at 60 secs my test audio had only 4 chunks and the PR branch is failing on a particular region of audio (is it related to dynamics?). So it's probably much better than 75%.

Also, I should have read the comments -

        roundoff (float): will round the pdf up to that level to remove difference coming
        from e.g. evaluating the Language Model on different architectures.

but having tried it Im not sure that works.