perf: use more rayon over tokio

[baszalmstra]

This PR replaces our tokio logic for IO with rayon yielding a significant performance improvement at least on Windows. But we need more testing on other operating systems.

This also incorporates the changes from #940.

> ./target/release/rattler_main create --platform win-64 jupyterlab -c https://prefix.dev/conda-forge 
...
48sec 220ms 578µs

With these changes:

> timeit { ./target/release/rattler create --platform win-64 jupyterlab -c https://prefix.dev/conda-forge }
...
2sec 620ms 702µs