Napari: Investigate and fix performance issues when drawing multiple layers

[jni]

@Czaki showed some profiling results on Zulip when adding "many" napari layers, specifically 100 images. It seems a lot of time is spent on computing the world extent:

so that's a good low-hanging fruit.

We do have benchmarks, so adding a benchmark for adding "many" small images would be useful, too.

[jni]

Many thanks to @Czaki for the investigation! 🙏

[Czaki]

A short tutorial for reproducing without using PyCharm Professional.

May be need to patch running_as_bundled_app

def running_as_bundled_app() -> bool:
    """Infer whether we are running as a briefcase bundle"""
    # https://github.com/beeware/briefcase/issues/412
    # https://github.com/beeware/briefcase/pull/425
    try:
        app_module = sys.modules['__main__'].__package__
        metadata = importlib_metadata.metadata(app_module)
        return 'Briefcase-Version' in metadata
    except Exception:
        return False

Profiled code in sample_code.py

import napari
import napari.plugins.io
import sys

with napari.gui_qt():
    viewer = napari.Viewer()
    for iteration in range(0,100):
        print(iteration)
        viewer.add_image(
            napari.plugins.io.read_data_with_plugins("image.png")[0],
        )
    viewer.close()
    sys.exit()

Install graphviz program
Install gpof2dot package in the python environment . Then call

$ python -m cProfile -o napari.pstat sample_code.py
...
$ python -m gprof2dot -f pstats napari.pstat | dot -Tpng -o output.png

Here sample output (with removed nodes which total time bellow 6% because of github limitations)

[sofroniewn]

Just noting that this has been dramatically improved with the merge of #1945, but we still have #1949 and #1950 in progress that could cause further improvements.