docs: collect the papers and standards the calculations are based on

[HuggeK]

The calculations in core and the roofmodel module rest on published papers,
standards and prior work, and today those references live only as scattered
code comments. Someone auditing the power math, or a reviewer asking "where
does this correlation come from?", has to grep for them. The docs tree should
carry one small page — say docs/references.md — that lists each
calculation's source next to where it is used, the same way
docs/site-convention.md
anchors the sign convention.

Rules to keep it useful:

• One line per reference: what it is, what FTW uses it for, and the code
  entry point that implements it.
• Only sources a calculation actually follows — no further-reading lists.
• New PRs that implement a published method add their reference in the same
  change (reviewable in the diff, like a changeset).

Draft inventory from the current code comments (starting point, not exhaustive)

Source	Used for	Where
Reda & Andreas (2003), Solar Position Algorithm for Solar Radiation Applications, NREL/TP-560-34302	reference for the solar-position approach	go/internal/sunpos
Spencer (1971) Fourier series (declination, equation of time, orbital distance)	the simplified solar position actually implemented (~0.05°)	go/internal/sunpos
Erbs, Klein & Duffie (1982) diffuse-fraction correlation	GHI → DHI/DNI split when a provider publishes only global horizontal	go/internal/sunpos, used by the forecast POA projection
WMO sunshine-duration threshold (120 W/m² direct beam)	deriving cloudiness from STRÅNG sunshine minutes	go/internal/strang
Yavuzdoğan (2023), Renewable Energy, doi:10.1016/j.renene.2023.119022 (the SPAN method)	iterative RANSAC + DBSCAN roof-plane segmentation (method reimplemented; no code taken from the GPL plugin)	roofmodel/ftw_roofmodel/segment.py
Lantmäteriet's Gauss conformal projection (Krüger series) publication	direct SWEREF 99 TM ↔ WGS84 conversion instead of a pyproj dependency	roofmodel/ftw_roofmodel/sweref.py
OGC 12-128r19 GeoPackage Encoding Standard, clause 2.1.3	decoding building-footprint GeoPackages without GDAL	roofmodel/ftw_roofmodel/geopackage.py
OGC 06-103r4 Simple Features, clause 8.2 (WKB)	geometry parsing inside those GeoPackages	roofmodel/ftw_roofmodel/geopackage.py
COPC specification (copc.io) / ASPRS LAS 1.4	range-request windowed point-cloud reads	roofmodel/ftw_roofmodel/pointcloud.py
STAC API specification	catalog search contract (POST {base}/search), Lantmäteriet and any custom catalog	roofmodel/ftw_roofmodel/geotorget.py
Recursive least squares (standard formulation) + Kalman smoothing	the self-learning PV twin and load model	go/internal/pvmodel, go/internal/loadmodel, go/internal/telemetry
Median / median-absolute-deviation robust statistics	the STRÅNG performance-calibration factor that one snowy day must not move	go/internal/pvperf

If the PV-forecast calibration work grows into a documented method of its own
(physical prior + statistical correction), the forecasting literature behind
that choice belongs on the same page — but only once code actually follows
it.