Add mechanism to report warnings/anomalies during extraction

[tidoust]

Extractors in browserlib cannot report curation actions that they may perform in order not to generated invalid data. Only things they can do is:

1. Report warnings and errors to the console. No one monitors these messages in production.
2. Throw an exception and make the extraction fail. That's appropriate in some cases but generally not what you want.

One idea could be to make extractors return a more structured JSON response that e.g. includes "warnings" and "errors" properties, which could then be picked up by Strudy, test code and/or other tools to alert us that a spec has unexpected content that probably needs looking into.

Via #1110 (adding a curation step for dfns does not remove the need for such a mechanism)