I can't really be bothered to take a look at the tests, but I strongly doubt there are actually 500 features. A large part of those tests are probably trying combinations of features. E.g. suppose markdown
only had tables as a feature, and nothing else. That feature alone deserves several several tests (for tables of various sizes, edge cases such as having
only the header, having rows with an incorrect number of columns, etc.).
But let's assume we can get away with just a single test for tables. And then we introduce the features "section headers" and "bold" and "underline". All these features can interact (e.g. underlined bold section headers), so we want to test combinations of all those features, and have a nice combinatorial explosion.