fix(itkwasm): fall back to the image size when the buffered region is empty

[vboussot]

Summary

shape = list(image.bufferedRegion.size or image.size)[::-1]

Context

#1545 switched the reshape source from size to bufferedRegion.size, which
information-only outputs need since they carry a logical size but no allocated
buffer.

Not every producer fills the buffered region though. Readers in
itkwasm-image-io 1.6.1 return it empty, so the reshape target became an
empty tuple and every read fails:

from itkwasm_image_io import imread
imread("cthead1.png")
# ValueError: cannot reshape array of size 65536 into shape ()

1.0b195 + image-io 1.6.1  ->  OK,   bufferedRegion=[]
1.0b199 + image-io 1.6.1  ->  ValueError

The change landed on 30 June but only surfaced when 1.0b199 shipped, since
nothing was published in between.

Two producers use two conventions, and one line has to serve both:

producer	size	bufferedRegion	correct source
information-only output	[10,34,34]	[0,0,0]	buffered region
file reader	[256,256]	[]	size

or covers both without giving up either: [0,0,0] is a non-empty list and
still wins, [] falls back.

Validation

test_pipeline_image_without_buffered_region mirrors the information-only
test from #1545 for the empty case. It fails without this change and passes
with it, and the information-only test keeps passing either way.

Core Python suite: 35 passed, 4 skipped, excluding the Pyodide and JS package
config tests.

Downstream, ngff-zarr went from 23 CI failures to 811 passed against a patched
1.0b199, and reading a file then downsampling it through
itkwasm-downsample-cucim works end to end on a GPU.

Also worth a look

pyodide.py has a fallback but tests is not None, so it would miss this
case, which is a buffered region that is present with an empty size. Same
latent issue, same fix. I have not exercised that path.

[vboussot]

@thewtex The remaining red check is main's red, not this PR's: itkwasm-python (ubuntu-24.04, 12, compare-images) runs the downstream packages' Pyodide tests against itkwasm 1.0b199 from PyPI. The same single job fails on main's latest Python Wasm run. It will turn green once this lands and itkwasm is republished; every job that exercises the branch's code passes.