Survey-aware light-curve column auto-detection

[tjayasinghe]

Summary

Makes column auto-detection plug-and-play with real light-curve downloads from the
major time-domain surveys — cup.periodogram(df, "GLS") works with no column hints
for ASAS-SN, ASAS-3, ATLAS, CRTS/CSS, ZTF, Pan-STARRS, LSST, TESS, Kepler, Gaia, and
MACHO. Headers were taken from each survey's authoritative format docs (IRSA, MAST,
ESA Gaia datamodel, LSST sdm_schemas, fallingstar-data, ASAS/OGLE READMEs).

What changed

• Expanded detection lists (columns.py) with the real spellings each survey uses —
  e.g. magerr (ZTF/CRTS), flux_error (Gaia), uJy/duJy/dm (ATLAS),
  psfFlux/psfFluxErr/obsTime/filterID (Pan-STARRS), midpointMjdTai/band +
  DP0.2 midPointTai/psFlux/filterName (LSST), MAG_0/MER_0 (ASAS-3),
  filtercode/HMJD (ZTF), KSPSAP_FLUX (TESS QLP), bkjd (Kepler).
• Domain- & measurement-aware error pairing. The error is derived from the resolved
  value column first (PDCSAP_FLUX→PDCSAP_FLUX_ERR, uJy→duJy), then the
  domain-appropriate list, then neutral err/sigma. Keeps a flux value with a flux
  error and a magnitude value with a magnitude error even when a table has both (ATLAS),
  and avoids the ATLAS F-filter vs flux-f collision.
• Corrected/detrended fluxes preferred (PDCSAP_FLUX, KSPSAP_FLUX) over raw
  SAP_FLUX; infer_domain recognizes flux units (uJy/mJy/nmgy).

Docs & tests

• A Supported surveys table in the light-curves guide, with a note on headerless
  products (OGLE .dat) and positional reading.
• A parametrized test over 13 real survey header sets asserting correct
  time/value/error/band resolution (incl. the ATLAS and TESS pairing subtleties).

Verification

pytest 137 passed / 2 skipped · ruff + mypy clean · docs -W build clean.

Limits (documented)

• ATLAS's single-char F filter, and OGLE/MACHO headerless or undocumented spellings,
  need an explicit ColumnMap.
• Gaia's wide TAP table (g_transit_flux…) is out of scope; the standard flat epoch
  product is covered.

🤖 Generated with Claude Code