astrometric_uncertainty_Veres2017 selects on catalogue names, but obs80 supplies single-char codes

[matthewholman]

Summary

astrometric_uncertainty_Veres2017 selects its per-station σ by comparing catalog against
catalogue names ("Gaia1", "UCAC4", "PPMXL", …), but Obs80Reader supplies the
single-character MPC code ("U", "q", "t", …). Every name-keyed branch is therefore
unreachable for obs80 input, and the affected stations silently fall through to their else
value.

debias() already handles both spellings — that was the #409 fix, and its comment says so
explicitly:

catalog may be a catalog NAME (an MPC_CATALOGS key, e.g. "UCAC4", as ADES provides) or
the single-char CODE that obs80 supplies (an MPC_CATALOGS value, e.g. "q"). bias_dict is
keyed by the code, so map a name to its code and let a code pass through -- otherwise
debiasing silently no-ops on all obs80 input (issue #409).

astrometric_uncertainty_Veres2017 never got the same treatment.

Where

• src/layup/utilities/astrometric_uncertainty.py:87,107,117 and the 568 / Y28 / G83 /
  309 branches — all compare against MPC_CATALOGS keys.
• src/layup/utilities/file_io/Obs80Reader.py — "cat": slice(71, 72), a single character.
• src/layup/utilities/debiasing.py:79-88 — the correct handling, for comparison.

Reproduction

from layup.utilities.astrometric_uncertainty import astrometric_uncertainty_Veres2017 as V
jd = 2458000.0
for stn, code, name in [("568", "U", "Gaia1"), ("T09", "W", "Gaia3"),
                        ("Y28", "t", "PPMXL"), ("309", "q", "UCAC4")]:
    prg = "&" if stn == "309" else None
    print(stn, V(obsCode=stn, jd_tdb=jd, catalog=code, program=prg),
               V(obsCode=stn, jd_tdb=jd, catalog=name, program=prg))

568 1.5 0.1
T09 1.5 0.1
Y28 1.5 0.3
309 1.5 0.3

So a Gaia-referenced observation from 568 is weighted at 1.5″ instead of 0.1″ — 15× too loose,
225× in the weight.

Impact

Small in practice, and I would rather report that than overstate it. Measured over a 2.5%
shard sample of the MPC archive as ingested for a full-catalogue fit:

	observations sampled	σ changes if the code is decoded	mean σ as-is	mean σ decoded
numbered	13,272,316	3,345 (0.025%)	0.7887″	0.7884″
unnumbered	623,714	1,297 (0.208%)	0.5946″	0.5922″

Station 568 dominates the affected set. The reason the exposure is so low is that the two
largest catalogue classes in the archive — Gaia DR2 (V) and Gaia EDR3 (X), together 43% of
numbered and 39% of unnumbered observations — are not in MPC_CATALOGS at all, so they take
the generic elif catalog: 1.0 branch either way.

It did not move any result we depend on. It is still wrong, and it will matter more as the
Gaia-referenced fraction from those specific stations grows.

Suggested fix

Mirror debias() — normalise the input before the comparisons:

from layup.utilities.debiasing import MPC_CATALOGS
_CODE_TO_NAME = {v: k for k, v in MPC_CATALOGS.items()}
catalog = _CODE_TO_NAME.get(catalog, catalog)   # code -> name; a name passes through

Two things worth deciding at the same time:

1. MPC_CATALOGS has no entry for Gaia DR2 (V), Gaia EDR3 (X), UCAC-5 (Y) or ATLAS-2
   (Z). For debiasing that is correct — the Farnocchia tables have no column for them and
   Gaia-referenced astrometry needs no correction — but the Veres model does want to
   distinguish them, so the σ path needs its own code→name map rather than borrowing the
   debiasing one.
2. "Gaia3E" appears in the σ branches but not in MPC_CATALOGS, so the two modules disagree
   on the naming already.

Version

main @ b6d7edd.

_{Surfaced while reconstructing how a full-MPC catalogue fit was weighted.}