In some cases the render result for a layer may depend on more than
one layer.
This changes QgsMapRendererCache to use "cache keys" instead of layer ids
to identify images within the cache. When setting an image inside the
cache, a new list parameter is used to specify all layers on which
that image depends.
Rationale:
- there was a lot of large objects passed by value, so potentially
there's a speed bump from this
- even for implicitly shared classes like QString/QList there's still
a (small) cost for copying the objects when there's no reason to
- it's the right thing to do!
