Fix globular clusters mis-typed as galaxies in NGC/IC import

[brickbots]

Problem

A user reported NGC 7006 is listed as a galaxy but is a globular cluster. Investigation showed it's not alone — five globulars are mis-typed as Gx:

Object	Steinicke TYPE	Was	Now
NGC 2808	I	Gx	Gb
NGC 5824	I	Gx	Gb
NGC 5834	I	Gx	Gb
NGC 6864 (M 75)	I	Gx	Gb
NGC 7006	I	Gx	Gb

Root cause

For a globular, Steinicke's TYPE is the Shapley–Sawyer concentration class — a Roman numeral such as I — which is_galaxy_type() read as an Irregular galaxy. preprocess_steinicke_type() already had a rescue that reclassifies these as Gb when a GCL/globular cross-ID is present in the remarks, but it ran after the galaxy check, so it was dead code for exactly these objects.

Fix

Move the globular disambiguation ahead of the galaxy check. The existing is_trumpler_class() guard ensures objects that merely mention a globular in their remarks are not reclassified — e.g. IC 4802, a star group inside Pal 9, correctly stays Ast.

Changes

• steinicke_loader.py — reorder the globular rescue before is_galaxy_type, with an explanatory comment.
• test_catalog_data.py — add NGC 7006 and NGC 6864 (M 75) as Gb regression assertions.
• pifinder_objects.db — regenerated via the full import pipeline.

Verification

All 4 tests in test_catalog_data.py pass; the rebuilt DB shows the 5 globulars as Gb and IC 4802 as Ast.

🤖 Generated with Claude Code

Test coverage (why this slipped through)

The previous tests only spot-checked a handful of famous objects in the built DB, and test_object_counts is blind to type errors (Gx vs Gb is still one object). Added python/tests/test_steinicke_parsing.py:

• Unit tests of preprocess_steinicke_type() — one case per decision branch and known ambiguity (Roman-numeral galaxy/globular/Trumpler overlap, combinations, suffix stripping, the IC 4802 guard). No DB required.
• An invariant audit over the real Steinicke source: every object with a globular cross-ID and a concentration-class TYPE must map to Gb — catching the whole class generically rather than enumerating named objects.

31 new tests; full suite (test_steinicke_parsing.py + test_catalog_data.py) green.