Blobdetector gives incorrect rounds and channels for spots when run with no reference image and is_volume=False

[shachafl]

Description

Spots results from starfish/core/spots/FindSpots/blob.py are saved with incorrect round and channels when using no reference image and "is_volume=False" (either a 2d data or 3d data that is processed as a set of 2d tiles).

Working on #2154 I also discovered a different issue with different results between "is_volume=True" and "is_volume=False" when no reference image is used. It is due to the way spots are merged from the same round/channel but different z slices:

starfish/starfish/core/spots/FindSpots/blob.py

Lines 210 to 233 in 991aded

	# If not a volume, merge spots from the same round/channel but different z slices
	if not self.is_volume:
	merged_z_tables = defaultdict(pd.DataFrame) # type: ignore
	for i in range(len(spot_attributes_list)):
	spot_attributes_list[i][0].spot_attrs.data['z'] = \
	spot_attributes_list[i][1]['z']
	r = spot_attributes_list[i][1][Axes.ROUND]
	ch = spot_attributes_list[i][1][Axes.CH]
	merged_z_tables[(r, ch)] = pd.concat(
	[merged_z_tables[(r, ch)], spot_attributes_list[i][0].spot_attrs.data])
	new = []
	r_chs = sorted([*merged_z_tables])
	selectors = list(image_stack._iter_axes({Axes.ROUND, Axes.CH}))
	for i, (r, ch) in enumerate(r_chs):
	merged_z_tables[(r, ch)]['spot_id'] = range(len(merged_z_tables[(r, ch)]))
	spot_attrs = SpotAttributes(merged_z_tables[(r, ch)].reset_index(drop=True))
	new.append((PerImageSliceSpotResults(spot_attrs=spot_attrs, extras=None),
	selectors[i]))
	spot_attributes_list = new
	results = SpotFindingResults(imagestack_coords=image_stack.xarray.coords,
	log=image_stack.log,
	spot_attributes_list=spot_attributes_list)

Bottom line, the {round,ch}I pairs are saved incorrectly ({0,1} should be {1,0}, and so on).

Steps/Code to Reproduce

Run the script examples/how_to/assess_spotfindingresults.py, without a reference image and once with "is_volume=True" and once with "is_volume=False".

# run blob detector on dots (reference image with every spot)
bd = FindSpots.BlobDetector(
    min_sigma=1,
    max_sigma=3,
    num_sigma=10,
    threshold=0.01,
    #is_volume=False,
    is_volume=True, ## DEBUG1 - True instead of False
    measurement_type='mean',
)
#spots = bd.run(image_stack=imgs, reference_image=dots)
spots = bd.run(image_stack=imgs)  ## DEBUG2 - No reference image

# Start Debugging
for r in range(imgs.num_rounds):
    for ch in range(imgs.num_chs):
        data=spots[{Axes.CH:ch, Axes.ROUND:r}].spot_attrs.data
        print(f"\nRound:{r}, Ch:{ch}")
        print(data.sort_values("intensity", ascending=False).head(5))
# End Debugging

Expected Results

The same results for "is_volume=True" and "is_volume=False" when the same data is presented as 2d (y,x) or 3d with a single z-plane (1,y,x), especially as we now squeeze the single z-plane and send blob_log() the same 2d data.

Actual Results

One example result for the case of "is_volume=True":

Round:0, Ch:1
     intensity  z    y    x  radius  spot_id
217   0.225376  0  938  428     2.0      217
13    0.223259  0  890  344     2.0       13
115   0.222446  0  158  979     2.0      115
36    0.221881  0  309  586     2.0       36
70    0.221012  0  446  212     3.0       70

One example result for the case of "is_volume=False":

Round:0, Ch:1
    intensity  z    y     x  radius  spot_id
0    0.196602  0  472  1174     2.0        0
1    0.188165  0  452  1093     2.0        1
15   0.186682  0  944   338     2.0       15
10   0.185512  0  549   782     2.0       10
34   0.184856  0    7   980     2.0       34

[shachafl]

close with #2160