Also, I would prefer having an option to ignore all errors, e.g. with option ERROR_LIMIT set to -1. Because it is rather difficult to estimate a number of future errors if you are playing with some badly structured data, while always setting it to 100500k looks ugly.
Here are the patch that contain all the comment given except adding a way to specify
to ignoring all error because specifying a highest number can do the work and may be
try to store such badly structure data is a bad idea