GitLab

So that a caller interested in in the ID doesn't have to fetch the
entire entity

All request that could return more than one row (ie. all member function
that used to return an std::vector<std::shared_ptr<T>> will now return a std::unique_ptr<Query<T>>
The query object can return:
- The amount of row represented by that query (->count())
- A subset of the rows (->items(count, offset))
- All rows (->all())

Which was created as is_track_presentAFTER, which apparently defaults to
a BEFORE trigger, causing the wrong is_present value to be used

Automatically delete genres with no tracks

It's more reliable than using the media as duration source, since the
media might not exist when reacting to a track disappearance
It also speeds up a bit the album modification triggers since we don't
need an inner select to get the added/deleted track duration

See https://www.sqlite.org/foreignkeys.html#fk_indexes
Having an index on any column that is a foreign key is advisable (if not
strongly recommended)

So we avoid getting bitten by ODR

fix cid #65548

This allows us to fetch more resources, and will prevent obscure bugs if
the connection is reopened, as the old pointer wouldn't be valid

AlbumTracks now store a genre id, instead of a plaintext genre.

This also fixes an assertion when refusing to fetch an entity because of
a failed precondition, since we wouldn't cache the entity.
But moreover, it reduces contention

Instead, have a FOREIGN KEY in those extension tables
This also mandates a media to be provided when creating a
movie/showepisode, and constify the extension getters

This is far from complete, but it's a start

Instead of storing a simple string