More precise and adaptive JSON parsing speed prediction

[atteneder]

In order to avoid frame stalls, glTFast predicts the time it takes to parse the glTF JSON and depending on that it parses on the main thread directly or moves it to a worker thread.

This prediction is based on a throughput measurement taken years ago (so with different JSON parsing code) on on particular machine (the representation for all cases is questionable).

While running performance tests I observed major frame stalls because the expected parsing time was totally underestimated (on the same machine the original measurements were taken).

We should:

• Make better educated guesses pertaining to the expected speed (e.g. lower on mobile platforms).
• Measure the actual speed and adjust the expected throughput. That should lead to better estimates on subsequent loads.
• Override the estimation in GLTFast.Newtonsoft.GltfImport, as Newtonsoft JSON has different performance.