Evaluate error recon approach

[Florian Obermayer]

Currently, everything what cannot be parsed will turn into a comment. This is bad for linting!

2 approaches to evaluate:

1. redesign comment grammar (instead of inversive thinking, directly express comment as markdown)
2. statefull parsing: monitor entering and leaving "sensible" blocks, so when block was entered successful but not left, assume there is an malformed string inside

[Florian Obermayer]

implemented an error tracing sytem where you can publish errors (with locations) and also track sensible blocks (with locations)