Refine MatchSpec validation to avoid incompatibilities with libmamba's parser

[jaimergp]

Description

Closes #418

We were a bit too strict with some auto-generated MatchSpec. The issue above includes an environment where a user has installed a package via URL or path. In this case, conda generates a MatchSpec object that gets rendered as package-name[url="package-url"].

libmamba only supports some key values in the brackets, so we were rejecting url with a InvalidSpecError. However, url here is superfluous so we can simply drop it from the passed string.

This uncovered a new matchspec that makes libmamba segfault: <unknown>/noarch::test-package. We will also anticipate that and drop it before the string is generated because, again, it doesn't add any useful info for the solver and it's only an artifact of not knowing from which channel test-package was installed (because it came from a URL).

Note that if the URL happens to be a channel URL (e.g. conda.anaconda.org/conda-forge/noarch/package-version-build.tar.bz2), conda DOES recognize it as a valid channel and that doesn't crash. It only reproduces if the tarball is in a non-channel-like location.

Checklist - did you ...

• Add a file to the news directory (using the template) for the next release's release notes?
• Add / update necessary tests?
• Add / update outdated documentation?

[jaimergp]

Failing as expected. Submitting fix now.

[therc]

Would this also fix the error

InvalidMatchSpec: Invalid spec 'conda-forge/linux-64::_libgcc_mutex==0.1=conda_forge[md5=d7c89558ba9fa0495403155b64376d81]': Libmamba only supports a subset of the MatchSpec interface for now. You can only use ('name', 'version', 'build', 'channel', 'subdir'), but you tried to use ('md5',).

assuming that "md5" would/could be dropped, too?

This was while trying to install another unrelated package. _libgcc_mutex shows as

_libgcc_mutex 0.1 conda_forge conda-forge

and was originally installed through a lockfile (which is where I suspect md5 comes from).

[jaimergp]

Hm, maybe not (yet). We are only dropping the url field. Can you open a new issue so we can take a look? It's weird that the md5 hash ends up in the matchspec.

[nhsavage]

new issue opened at #427