Collect errors instead of failing early

[sschuberth]

This continues the discussion from here. Instead of failing early / fast e.g. on encoding errors, IMO it would be nice to continue checking remaining files, in order to see how many errors there are in total and / or bulk fix all errors instead of fixing them incrementally.

Both failing early and collecting errors have their valid use-cases. E.g. on CI you sometimes want to fail early to reduce feedback times. However, as lychee runs fast anyway, I feel there's little need to fail early to safe time.