Move to error interception rather than length checking

From https://github.com/cutenode/good-first-issue/pull/80#discussion_r239506631:

> If at this point we were to intercept an empty string literal from the response header rather than the results of a list, and this string literal was to be an error response in a short-circuit and no error was able to be parsed, edge of timeouts, parsing errors etc. This would still return as true, hence returning an empty string exposed.
It would be suitable at some point here validate the actual type integrity of the expected data formats being used, as side effects of this may lead to false positives.

Referring to [bin/good-first-issue#32](https://github.com/cutenode/good-first-issue/blob/master/bin/good-first-issue.js#L32).

Read bin/good-first-issue.js around line 32 and the referenced discussion in pull request 80. Clarify how response-header errors and unparseable or timeout responses should be intercepted, then verify that valid data is not mistaken for success or exposed as an empty string.

Written by the indexing model from the issue text.