Some ideas for extra heuristics

from a chat with @maxhbr

- a short license reference (single word?) detected after several 100 lines (say after the 1000's line is likely false positive

- a short license reference (single word?) detected in a binary and where the actual text has a different case than the case of the rule is likely false positive. For instance `gPL` or `gPl`. This could be also things that are done directly in Scancode toolkit