Always sort installed records and solver specs

[jaimergp]

Description

I've been thinking about some failures we get (only!) sometimes in CI, like tests/core/test_solve.py::test_conda_downgrade[libmamba] (see #317). These errors often go away with a retry in the same PR, so there's some randomness at play.

The hypothesis here is that some container is getting a different order depending on the attempt. SolverInputState.installed takes whatever order appears in PrefixData._prefix_records, which blindly "takes" whatever os.scandir() is returning. This function does not guarantee order, so we are going to switch to a toposorted installed container. Maybe this makes the tests behave more reliably?

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• Add / update outdated documentation?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]

Seems to work. The errors we saw in the last run (first attempt) are unrelated (#334, and a pip issue with classic that smells like rm_rf race condition).

See #317 for the whole debugging story.