Assess should classify failures to build/analyze packages

[tedinski]

When building a lot of packages, assess can fail for a variety of reasons. It'd be great to summarize these and identify the causes.

These include:

1. Kani-compiler panics, which we can aggregate based on the file:line of the panic
2. Build failures, where we can both have a "generic" build failure, and ad-hoc recognizers for specific known kinds of failure. (e.g. recognizing "assert is ambiguous" from futures-0.1 as a common specific reason. Building these ad-hoc recognizers might be "unprincipled" in that we just build them for everything we see in practice, but I think that's a good way to proceed: make sure we understand everything because we'll have a metric for "unknown build failure" that should trend towards 0.)
3. Common types of project misconfigurations. E.g. sometimes a native library development package needs to be installed on the host. Or projects can't build with their default features.
4. CBMC / goto-cc / validate-goto-model failures.
5. Other unknown failures.

[celinval]

Can we also store more information about the problem on the metadata? For example, the stack trace of a crash.