Fix unit handling and restore deleted logic in ScreenGalHalo image methods

[mandarmulherkar]

Note on stacking: this branch is built on top of #55, so it contains those commits too. Either approach works for me:

• Merge this one and close Fix unit handling in Halo.fake_image #55, everything is here (3 files, +95/−23); or
• Merge Fix unit handling in Halo.fake_image #55 first, and I'll rebase this onto main, leaving only the galhalo.py changes.

Happy to split it differently if you'd prefer.

What is broken

ScreenGalHalo.fake_variable_image, ScreenGalHalo.variable_profile, and the module-level time_delay all raise on every call. Six distinct defects, from two unrelated causes:

From the astropy-Quantity migration (949a0b6, June 2022):

• time_delay: d_cm = dkpc * 1.e3 * u.pc.to('cm') is a bare float, because Unit.to() returns a conversion factor rather than a Quantity. Dividing by c.c yields s/m, so .to('s') raises UnitConversionError for all inputs. time_delay is exported in all, so this is broken public API.
• variable_profile: passes self.theta (a Quantity) to time_delay, which documents alpha as a bare float in arcsec; the comparison in _is_small_angle then fails.
• variable_profile: the intensity accumulator is a plain ndarray while the values written into it carry 1/arcsec², raising TypeError.
• fake_variable_image: reads self.lam_unit, removed in that migration (same as Fix unit handling in Halo.fake_image #55 ).
• fake_variable_image: lmin/lmax compared against self.lam without units (same as Fix unit handling in Halo.fake_image #55 ).

From a docstring-only commit (fe4c562, June 2026):

• fake_variable_image: the line computing var_profile was deleted, leaving the loop below referencing an undefined name. Its trailing comment (# intensity cube (NE x NTH)...) was left behind, orphaned.

Why it went unnoticed

No test or notebook exercises any of these methods. 949a0b6 also deleted the only assertions that referenced lam_unit, so the suite stayed green while the attribute disappeared.

The fix

self.lam and self.theta carry their own units, so the unit-string branching is unnecessary. One .to(...) with the appropriate equivalency handles any input unit, and .value strips to a plain array where scipy needs one. lmin/lmax are given lam's unit before comparison. time_delay lets astropy do the kpc→s conversion instead of hand-rolling it. The var_profile line is restored; the time_now block deleted alongside it is not needed, since variable_profile already defaults tnow to time[-1].

variable_profile now returns a Quantity; its docstring is updated to match.

Tests

First coverage for all of these. 159 → 165 passing. Reverting any individual fix turns its corresponding test red. Verified one at a time, not just in aggregate.

Noticed but deliberately not addressed

• path_diff and time_delay are public and would be friendlier accepting Quantities, since halo.theta is one. That's an API change rather than a bug fix, happy to follow up.
• theta_rad (galhalo.py:171) and lctm (:175) are assigned and never read.
• deltat is computed inside a loop it doesn't depend on.
• iend = max(...) looks like it drops the bin at lmax, in both fake_image and fake_variable_image.
• The intensity accumulator assumes self.fabs is unitless; a Quantity flux passed to calculate_intensity would still fail.

Co-authored by Claude: Claude explained the physics and did the code reviews; I wrote the tests and code fixes.