sextractor_runner runs dual-image mode with the same image twice, changing measurements

[cailmdaley]

sextractor_runner always calls SExtractor in dual-image mode (source-extractor img,img -WEIGHT_IMAGE w,w), even when DETECTION_IMAGE = False and the two images are the same file. On UNIONS tile 186.307 (DR6 image and weight, default_tile.sex), this gives the same detections and positions as a single-image call, but different measurements.

Compared with the DR6 catalogue, 20 < MAG_AUTO < 24.5	single-image call	dual-image call (the pipeline)
|ΔMAG_AUTO| > 1e-3	0.12%	22%
|ΔMAG_AUTO| > 0.01	0.03%	3.7%
FLUX_RADIUS differs	0.11%	20%

BACKGROUND differs for 28% of objects. XWIN moves by 1.6e-3 px at p90 and 0.024 px at p99. I haven't traced which part of SExtractor causes this; the background estimate is the obvious suspect.

Consequences.

• Data and image sims both run the runner, so they are consistent with each other.
• The tile measurements are not the ones in the UNIONS catalogue, although the configuration is the same (Check that sim-side tile SExtractor reproduces the DR6 detections data uses #929). The photometry catalogue and the shape catalogue therefore differ at the 1e-2 mag level for about 4% of objects.
• The exposure-side SExtractor that selects PSF stars uses the same call.

Proposal: call SExtractor with a single image when DETECTION_IMAGE = False. That changes tile and exposure measurements, so it's a campaign boundary. It belongs after #933, which makes data run this SExtractor path.

Evidence: candide:/n17data/cdaley/scratch/xwin-assoc/e2e/modes/ (single.fits, dual.fits, logs) and ../dual_vs_single.py.

Claude Opus 5.5 on behalf of Cail