Fix unit handling in Halo.fake_image

[mandarmulherkar]

What is broken

• fake_image raised AttributeError on every call; two further errors were masked behind it.
• Root cause: self.lam_unit was removed in 949a0b6 (June 2022) during the astropy-Quantity migration; two reads of it survived in halo.py and galhalo.py. That commit also deleted the only tests/assertions referencing it, so nothing caught it.
• Why it went unnoticed: no test or notebook touches these methods.

Fix

• lam carries its own unit, so the unit-string branching is unnecessary. One .to(u.keV, equivalencies=u.spectral()) handles any input unit. lmin/lmax get lam's unit before comparison.

Tests

• First coverage for fake_image; each fails without its corresponding fix.
• Test count: 159 → 161 tests passing
• Reverting either fix independently turns the corresponding test red

Follow-ups (not addressed here)

• fake_variable_image in galhalo.py has the same lam_unit pattern, plus an undefined var_profile caused by commit fe4c562 ("remaking docstrings") deleting the line that computes it, and time_delay raises UnitConversionError for all inputs because d_cm is a bare float. Happy to follow up in a separate PR.
• iend = max(...) looks like it drops the bin at lmax; noted but deliberately left alone.

Co-authored by Claude: Claude explained the physics and checked my work; I added tests and code fixes.