Treat cabal file parsing failures from index as non-fatal

[snoyberg]

Some recent issues come from the fact that Stack has been (perhaps overzealously) too strict in its parsing logic. Given that we can't properly rely on upstream sources always providing input which is compatible with the deployed version of the Cabal file parser, we should relax this logic. (I made the original decision implicitly to be strict in parsing, due to past experiences with tools like packdeps and weird corner cases around BOMs, but this reasoning no longer applies.)

Here are some examples of things we should be less strict about:

• If a cabal file does not parse, ignore it. We need to figure out what that means for cases like a snapshot referring to a specific cabal file (perhaps disallow usage of that snapshot, perhaps drop that one package).
• Already implemented: don't add additional checks to flag name parsing: we'll just let in whatever the Cabal library accepts.
• Are there other data types that have additional checks not performed by Cabal?

We need to accept that there is no Cabal format specification, and as it stands now, the Cabal library implementation is the library.

[snoyberg]

Based on testing by @denibertovic, this is already implemented as needed. I'm going to move this out of the P0 milestone to avoid blocking the release, and will follow up with Deni on #3502 to get an integration test added.