fix: raw-flux latent + soft-fail magzero-required µJy

[Jammy2211]

Summary

total_galaxy_0_flux_mujy used to raise ValueError when its consumer enabled the
latent without passing magzero to AnalysisImaging. The raise fired post-fit
inside SearchUpdater._compute_latent_samples, killing the metric write of a
converged multi-hour search.

This PR adds a new raw-flux latent (total_galaxy_0_flux) that needs no
instrument inputs and ships default-on, and converts the existing _mujy
latent's raise into a NaN + once-per-process warning. The sibling change
for the lensing latents lives in PyAutoLabs/PyAutoLens#557 (linked once that PR is open).

API Changes

• Added total_galaxy_0_flux(fit, magzero=None, xp=np) — raw integrated
  flux of galaxy 0 in the fit's image units; no instrument inputs required.
  Default-on in autogalaxy/config/latent.yaml.
• Added module-level helper _maybe_magzero_warn(magzero, name) -> bool
  to dedupe the new soft-fail warning across the many fit evaluations in a
  search.
• Behaviour change on total_galaxy_0_flux_mujy(fit, magzero=None, xp=np):
  no longer raises ValueError when magzero is None — returns xp.nan and
  emits a single logging.WARNING per latent name per process.
• Default config change in autogalaxy/config/latent.yaml:
  total_galaxy_0_flux: true added; total_galaxy_0_flux_mujy stays false.

See full details below.

Test Plan

• python -m pytest test_autogalaxy/imaging/model/test_latent.py -v — 13/13 pass
• Full test_autogalaxy/ suite (subagent runs in CI commit)
• Smoke test impact on autogalaxy_workspace — verified there are no callers of the removed raise in workspace scripts (separate workspace PR will demo the new latent in guides/units/flux.py)

Refs PyAutoLabs/PyAutoLens#556

Full API Changes (for automation & release notes)

Added

• autogalaxy.imaging.model.latent.total_galaxy_0_flux(fit, magzero=None, xp=np) — raw image-unit sum of fit.galaxy_image_dict[fit.galaxies[0]].array. magzero accepted but ignored (uniform dispatcher context).
• autogalaxy.imaging.model.latent._maybe_magzero_warn(magzero, name) — module-private helper. Returns True (and logs one WARNING) when magzero is None; False otherwise.
• autogalaxy.imaging.model.latent._MAGZERO_WARNED — module-level set tracking which latent names have already warned in this process.

Removed

• autogalaxy.imaging.model.latent.total_galaxy_0_flux_mujy no longer raises ValueError("magzero must be passed ...") on magzero=None. Callers that previously caught this ValueError will instead see xp.nan + a logging.WARNING record.

Changed Behaviour

• total_galaxy_0_flux_mujy(fit, magzero=None, xp=np): same signature, but magzero=None now → (xp.nan, logger.warning(...)) instead of raise.
• LATENT_FUNCTIONS registry now has 2 keys: total_galaxy_0_flux, total_galaxy_0_flux_mujy.

Default Config

• autogalaxy/config/latent.yaml gains total_galaxy_0_flux: true. total_galaxy_0_flux_mujy: false is preserved.

Migration

• No migration required for code that passes magzero (Euclid pipeline pattern unchanged).
• Code that previously caught the ValueError: drop the catch; the value is now NaN.
• Workspaces that want the µJy column auto-populated: enable total_galaxy_0_flux_mujy: true in their config/latent.yaml override AND construct AnalysisImaging(..., magzero=<value>).

🤖 Generated with Claude Code

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace#214

[Jammy2211]

Workspace PR: PyAutoLabs/autogalaxy_workspace#108

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace_test#133

[Jammy2211]

Workspace PR: PyAutoLabs/euclid_strong_lens_modeling_pipeline#19