Merge "snl" with "ndvi-snl" if geometry is the same

[bmarcos]

If the geometry of the "snl" and "ndvi-snl" Collections is the same, there should be only one (merged) Collection, since the fields in the attribute tables should be directly compatible.

[JappeF]

i agree, see #18 (comment)

[JappeF]

@bmarcos i fully understand the reasoning from a GIS attribute-table perspective. The blocker is that the two collections have different cardinalities:

• snl → one STAC item per parcel (~264K items, geometry only, no time axis). Item id: snl-<ndc_id>.
• ndvi-snl → one STAC item per (parcel × observation date) (35M+ items, geometry + NDVI stats per date, datetime is the time axis). Item id: ndvi-snl-<YYYYMMDD>-<ndc_id>.

The NDVI collection already embeds the parcel geometry (it's looked up by ndc_id at ingestion time), but it's repeated for every observation date. So a 1:1 merge isn't possible without either:

1. Collapsing the time dimension — pick one NDVI observation per parcel and lose the rest. Not viable.
2. Dropping the standalone snl collection — keep only ndvi-snl since it contains the geometry. Possible, but bbox queries for "give me all parcels in this area" would have to scan 35M NDVI rows instead of 264K geometry rows, and clients that want parcel boundaries without NDVI data would have to filter.

Leaving open. Option 2 might become viable once we can address the bbox-scan performance (partitioning by parcel-id, or a pgSTAC view).