ENH rattler

[beckermr]

To do:

• understand and fix timeouts (xref: fix: use total duration of timedelta objects conda/rattler#711)
• figure out how to add run_constrains to the solver calls (xref: how do you feed entries from run_constrains to the solver function in py-rattler? conda/rattler#712)

closes #19

[wolfv]

This is cool. Please let us know if you find anything that's wrong with rattler :) 🚀

[beckermr]

@wolfv @jaimergp This PR is ready to merge. A few notes.

• I punted on the .solv files with mamba since rattler appears to be working with the test suite here. That is pretty non-trivial actually. The test suite here covers some very complicated feedstocks (arrow-cpp) and depends on properly interpreting run_constrains and virtual packages.
• I did not use the built-in rattler virtual package functionality and am instead writing them as a channel on disk. This means we can use the same code with mamba.
• The timeouts appear to be working with rattler.
• rattler appears to be caching repodata very well
• The timing for the test suite comes in with rattler about 10x faster. IDK if this is the solver, the metadata loading, or both.
• I disabled the mamba test suite and made it not required. When we added support for run_constrains to the solver, it increased the test suite run time from ~20 minutes to ~1 hour which is a lot.

[wolfv]

Thank you @beckermr for pushing this through. This is an awesome test for resolvo and it's good to know that it works!

Also happy that we can make use of the features that we added to resolvo especially for these cases (timeout etc.) :)