Move the calculation of the pre/post context to the search workers

The logic behind that is that it potentially uses less memory, but the
timing differences are negligible. Updating the match marks takes
almost the same amount of time as the actual searching in an open file.

The multi line post-context calculation is now also corrected.