feat: magnification latent from raw fluxes, no magzero (arc phase 7a) #812

Arc phase 7a (epic cluster-strong-lensing, phase 7 posterior-draw magnification errors). Rewire the magnification latent from the µJy latents (NaN without magzero) to the raw-flux latents (total_lensed_source_flux / total_source_flux), identical for positive fluxes since the µJy factor cancels. Keep it default false; document it only in config/latent.yaml and autolens_workspace guides/results/latent_variables.py (human ruling 2026-10-30: latents are explained in results only, not brought into every workspace script). Per-image point-magnification latents are not added (image count varies by dataset). Phase 7b (posterior-draw errors in every source_science.py) follows separately.