[alexrichey]

Follow-ups from the 26c triage of unaccounted diff rows. One output change: the exception table now skips names shorter than 2 characters, as prod's ExceptionExtractClass does, which drops "E". Everything else is QA accounting. SND, enders and exception rows prod has but our source doesn't cite CSCL-NORM-01 (by rule, plus 6 modified SND rows by key). Dev-only LDF G/P records are accounted for under CSCL-LDF-01 by a rule matching prod's IsSegmentdeleted, and a differing LDF header now counts as one row, accounted for when only its record count differs, by the LDF-01 net. gdb_lion SegCount off by 1 on a centerline with a same-node centerline within 0.0005 ft is cited under CSCL-LION-14. New seed row CSCL-DISTRICTS-07 covers nypp precincts that match once prod's parts and holes under 100 sq ft are removed, as ours drops them.

Verified on a private 26c DuckDB build, with the affected chains rebuilt and the rest deferred to the Oct 21 build. Unaccounted rows: snd 102 to 0, enders 6 to 0, exception 4 to 0, ldf_dat 12 to 0, ldf_header 2 to 0, gdb_nypp 11 to 5. gdb_lion SegCount 0133914/0161031 are now accounted for. 10 unaccounted rows are left in total: nyct2010(wi) 5, waiting on GR, and nypp 41, 72, 88, 110 and 113. Four of those nypp precincts come from zero-width spikes on our own outline, which will be fixed separately. The known_differences seed tests, the citation test and sqlfluff pass. A full CI build wasn't run locally. Tracker: inv-2cy.57.