fix COLP UID

what's wrong

geo_inputs.sql builds uid as MD5(CAST((dcas_ipis.*) AS text)) so it's a hash of the entire source row, including two columnds added during data loading (ogc_fid, data_library_version) that change every build

1. Every UID changes every release. Across 10 published versions (2021→2026), zero UIDs persist, though ~16–17k records persist from one version to the next
2. Duplicate rows SELECT DISTINCT in create_colp.sql includes uid, so identical source rows with different ogc_fid survive as duplicates (88 rows in 20260430, all originating in IPIS)

what we should probably do

• Base uid on an explicit list of columns to create a natural key, not .* to make it version-stable and to let SELECT DISTINCT drop the duplicates
  • should probably be BBL+AGENCY+USECODE+LEASED and generated during create_colp.sql
• Move deduplication to ingest or early in the build (the duplicates are an IPIS source data)
• Resolve the few conflict groups that won't deduplicate: 3 differ on owned/leased (O vs L, should probably remain distinct), 1 on u_f_use_code (already flattened downstream as finalcom='D')

Start with geo_inputs.sql and create_colp.sql, tracing how uid is built and how SELECT DISTINCT is applied. Check the IPIS-derived duplicate and conflict groups described in the issue, then define the explicit natural-key columns and deduplication point. Done means UIDs persist across releases and the identified duplicates are removed without incorrectly merging the owned/leased conflicts.