When we close the mnemosyne application and re-open it, it won't
remember the number of the already memorized cards. In other words, it
won't warn you if you've already learned 15 or more words.

This PR aims to fix this issue and re-construct the
`self._fact_ids_memorised` from the logs.

We can fetch all the newly learned cards and all the forgotten but
learned cards today. Combining the two, we can get the number of the
learned cards today.

When saving the Settings dialog, the `reset` is alwas called, so it 
would trigger an alert every time when, because during the reset 
function, the `warned_about_too_many_cards` is reset to 
`warned_about_too_many_cards = False`. 

This fix initializes the `warned_about_too_many_cards` on the class 
level, and calling  `reset` won`t reset it back to `False`.

Also only warn when we've just reached the 15 new or forgotten words.

Add a new event type: `WARNED_TOO_MANY_CARDS` so we can query from the 
log table if the warning has been already shown or not.
- Add test coverage

Warning on start-up would be only useful if we could alert more than 
once a day. If we already displayed a warning, we should never show the 
warning again on start-up.

Python lists are unordered, so comparing them will result in a random 
result.

Using `sort()` will ensure that the elements are ordered so we can 
compare them properly. Using sort is `O(n log n)`.

- Move sql queries to the database class
- Add test coverage for the new database API functions
- adjusting the variable naming to comply with the current naming scheme