Because some annotation items have scale dependent bounding boxes,
we can only index items which have a fixed bounding box. Other
item bounds need to be dynamically determined based on a specific
render context. So rework the annotation layer index handling
to only index appropriate items and store other items in a non-indexed
item set.
This isn't ideal, because it means we need to clone ALL non-indexed
items upfront whenever we render an annotation layer (it's too expensive
to calculate their actual bounding box and selectively clone them,
as it's a process which blocks the main thread).
Hopefully we can think of an alternative approach to this down
the line so that we DO have some form of spatial index for scale
dependent items, unlocking better performance for annotation layers
with 10,000s of items. But we'll ignore that situation for now ;)
to annotation layers to reflect that while these are editable
users are not able to toggle edits on them (they are ALWAYS
editable!)
Note that we can't use the existing QgsMapLayer::LayerFlag enum here,
as that has a different meaning (flags which are set by users
at runtime)
All the logic for rendering non-opaque layers is handled in a layer
type agnostic way already, so this just allows the existing handling
to be used for mesh/point cloud/vector tile/etc layers also
