Load all channels with conda internals

[jaimergp]

Description

Closes #59

---

We are replacing the libmamba networking stack with conda internals:

• We download all JSONs with the newest conda.SubdirData.repo_fetch interface that doesn't process the JSON (only downloads).
• We use libmambapy.Repo, without the SubdirData indirection. This interface allow us to use a JSON file directly (without having to be optimistic about the "find the JSON file" logic in SubdirData).
• All the tests that checked the libmamba networking stack have been removed, but the utilities are still there in case we ever need them in the future.
• These changes also allow us to drop tons of pre- and post-processing of MatchSpec and PackageRecord objects (e.g. percentage-encoding and decoding on local paths so libcurl doesn't choke, Windows things, etc). The libmamba<->conda index interface is better abstracted now.

---

There's some legacy issues we should remove in the future:

• The SubdirData subclass can introduce some issues with incomplete cache entries we have to invalidate (some tests break otherwise), because metaclass-based caches are that fun.
• In some tests, the cached files are not found because the integration tests abuse the SubdirData._cache_ a bit too much and some of those changes are not reflected on disk. Some changes upstream will alleviate this.
• TODO:
  • Change conda.testing.helpers._patch_for_local_exports to not patch _mtime to inf. Return early if the solver is classic because that function is not needed there.
  • Remove this block

[jaimergp]

Remaining errors have to do with conda.testing.helpers._patch_for_local_exports again. Since we are using SubdirData now, the cached entries get in the way of the testing mocking machinery.

[dholth]

We've merged a new conda that changes the way repodata.json is dowloaded, and we will change it again in the next release. Is this code compatible with the new conda? conda/conda#12050 also helps a lot avoiding wasted parsing.

[jaimergp]

Nope, conda/conda#12003 changed the internal API so as a matter of facts everything will fail here as it is now. I am inclined to wait til things get calmer around the repodata fetching logic before working on this PR further. What do you think @dholth?

[jezdez]

Nope, conda/conda#12003 changed the internal API so as a matter of facts everything will fail here as it is now. I am inclined to wait til things get calmer around the repodata fetching logic before working on this PR further. What do you think @dholth?

I concur, let's move this to a later milestone.

[jaimergp]

tests/test_solvers.py::TestLibMambaSolver::test_arch_preferred_over_noarch_when_otherwise_equal 

This one is failing sporadically. It can be reproduced locally by rerunning it several times; it will pass 4 out of 5 times or so. I wonder if this has to do with some change in 1.4.1 (this is the first time I see it fail, if memory serves).

[jaimergp]

I added some stdout in the spinner (so it hides when needed; e.g. --json or --quiet), and it looks like this:

During fetch:

$ CONDA_SUBDIR=win-64 conda create -n unused --dry-run python --solver=libmamba -c ../conda-src/tests/data/legacy_repodata/ -c defaults
Channels:
 - file:///opt/conda-src/tests/data/legacy_repodata
 - defaults
 - conda-forge
Platform: win-64
Collect all metadata (repodata.json): /

Once done:

$ CONDA_SUBDIR=win-64 conda create -n unused --dry-run python --solver=libmamba -c ../conda-src/tests/data/legacy_repodata/ -c defaults
Channels:
 - file:///opt/conda-src/tests/data/legacy_repodata
 - defaults
 - conda-forge
Platform: win-64
Collect all metadata (repodata.json): done
Solving environment: done

[jezdez]

Nice!