Apply debiasing for obs80 catalog codes, not just names (Closes #409)

[matthewholman]

Summary

debias() silently applied no correction on obs80 input: Obs80DataReader supplies the single-char catalog code (W, q, …), but debias() looked the catalog up as an MPC_CATALOGS key (a name like UCAC4). Since bias_dict is keyed by the code, a code never resolved → the guard returned the astrometry unchanged. So debiasing never fired on obs80 — including the bulk MPC archive the catalogue fit ingests.

Fix

catalog_key = MPC_CATALOGS.get(catalog, catalog)  # name -> code, or a code passes through
if catalog_key is None or catalog_key not in bias_dict:
    return ra, dec

• Name ("UCAC4") → its code "q" → in bias_dict ✓ (unchanged behaviour for ADES-style input)
• Code ("q") → passes through → in bias_dict ✓ (obs80 now debiases)
• blank / None / unknown → falls through the not in bias_dict guard unchanged ✓ (preserves debias() raises KeyError on blank/unknown star-catalog codes (crashes fits on real obs80 data) #401)

Validation

End-to-end, fitting the same objects with debias=False vs debias=True now differs (it was byte-identical before the fix):

obj	Δa (debias off)	Δa (debias on)
20000	3.83×10⁻⁴	2.16×10⁻⁴
90377	4.08×10⁻²	1.14×10⁻¹
944	5.27×10⁻⁸	6.89×10⁻⁸

Corrections are small (sub-arcsec), as expected — but they now actually apply, which the "faithful to JPL" weighting+debiasing story needs on obs80 data.

Tests

Adds a regression test asserting a code ("p") yields the same correction as its name ("PPM") and actually changes the astrometry. All 8 test_debias tests pass.

Closes #409.

🤖 Generated with Claude Code

[kjnapier]

Looks good