Fix database corruption in case of unexpected conditions

[scymtym]

In SBCL, evaluations resulting in type-errors or no-applicable-method-errors could result in database corruption because some slots of the conditions (:expected-type and :generic-function respectively) were printed unreadably, making subsequent loading of the database impossible.

The short-term solution, as implemented in the first commit, is adding the slots to stringified-error-prop?.

The second commit adds a "safety net", ensuring that values are actually printable readably.

Not sure what the correct solution is, but this is the fix I used.

[inaimathi]

Looks good as a short term fix.

The real solution for this is a tough call. It might ultimately be switching fact-base from (format s "~s~%" ...) and (read s) to something like a new back-end for cl-store. The default one has two problems with its output: it's binary, and it isn't trivially appendable. Maybe the first one of those doesn't matter too much, but two specific things we want to be able to do with data entries is

1. append new ones to the end of an existing data file, without re-writing the entire file
2. read existing entries from the end of an existing file

I'll give it more thought, but I'm open to suggestions.