I don't think we want to ignore all errors reading the ccache, though we
certainly do want to function as documented.  I think the code was
assuming that krb5_cc_default() would error out with KRB5_CC_NOTFOUND if
the cache didn't exist, and that's not the case with file ccaches.

Right now this is a little tough to fix cleanly because the file ccache
code returns a type-specific error (KRB5_FCC_NOFILE) when you try to
fetch something from an absent cache.  I'll need to figure out how we
want to address that.

(Also, if you're using this patch locally, I think you're using princ
uninitialized in the case where krb5_cc_default() returns KRB5_CC_NOTFOUND.)