Merge the per-dataset taxon shards into one authoritative reference

Each ingest emits the taxon rows its own vocabulary reaches. The same taxon can appear in several shards (Appendicularia is in both zoodb and zooscan), so rows are collapsed on taxon_key and each field takes the first non-NULL value in dataset directory order. There is no priority list (taxon plan D5): every shard's scientific_name / rank / classification comes from the same cached authority lineage, so shards agree wherever both have a value and the order only settles which shard fills a gap. common_name is not decided here at all — the release applies the written precedence with apply_taxon_common() — and notes is unioned, never picked.

Usage

merge_taxon_shards(
  con,
  root = ".",
  parquet_dir = cc_stage_path("parquet"),
  exclude = release_excluded_datasets(root)
)

Arguments

con: a DuckDB connection
root: workflows repo root
parquet_dir: directory holding the per-dataset output dirs. Defaults to the local staging root (see cc_stage_dir()), where the bulk parquet lives; an absolute path is used as-is, a relative one is resolved against root. The JSON sidecars stay in the repo and are found separately.
exclude: dataset dir names to skip (see core_shard_paths())

Value

(invisibly) the row count of the merged taxon