Better handle transient feed errors

[Altonss]

A lot of feeds might have transient errors, we should better handle it instead of opening new issues each time:

• Error fetching 'feeds/us-co.json' #767
• Error fetching 'feeds/ca-on.json' #768
• Error fetching 'feeds/us-oh.json' #769
• Error fetching 'feeds/it.json' #777
• Error fetching 'feeds/ca-bc.json' #781
• Error fetching 'feeds/us-nm.json' #783
• Error fetching 'feeds/us-il.json' #778
• Error fetching 'feeds/om.json' #790
• Error fetching 'feeds/hr.json' #793
• Error fetching 'feeds/ca.json' #780

[jbruechert]

I think this requires a redesign of the current issue reporting.

fetch.py could write the issues into a json file / sqlite database (per region file so we don't need to deal with locking).
It could also write things like exact error causes and count the number of identical error messages, timestamps etc.
Then in a second pass we could go over the generated reports in another script to make decisions what to open issues for.

Then for specific errors like timeouts, issues could only be opened after the issue occurs for more than a day or so.