reimporting is unintuitive

[martinpopel]

Background: If a new metric was added or its implementation was changed, the only way (I am aware of) how to update scores of already imported experiments is to re-import them (after git pull).
So you need to remove the experiments (and tasks) in question and import them again. You can remove the experiments with the "delete" button in the interface or (if you have disabled the button) manually. However, if you forget to delete the experiments from SQL and just delete the subdirectories of data, you will get strange error <TASK> has bad number of sentences when reimporting.

Solution: detect that we are importing a task which is already imported in the SQL database and provide a more informative error message than "bad number of sentences".