Directory file-count census

Extraction artifact, lane Extract.

What it contains

extract_directory_census.py — parser + re-usable extractor for the repository directory inventory (repo_tree.tsv).
CHECK — verification: re-derives the census from repo_tree.tsv and compares it against the committed tables + manifest.json. Exits 0 only if every check passes.
directory-census.tsv — ranked directory inventory (rank, directory, file_count).
namespace-buckets.tsv — namespace → directory/file-count aggregate.
directory-census.md — human-readable markdown table.
manifest.json — self-reported statistics (rows, total_files, top directory).

repo_tree.tsv carries a duplicated header row; both header lines are skipped.
A row is valid only when it has exactly three tab-separated fields and the rank + file-count parse as integers.
A row whose directory field starts with " is treated as malformed (it is the beam/CHECK file's content exported as a directory name, not a directory).
Rows are ranked by file_count descending; ties keep original rank order.

148 valid directories, 3343 total files, 1 malformed row.
Top directory: wizard-work/sphinx/archive-memgallery-2026-10-21/figures (635 files).
Top-25 directories account for 2638 files (~79% of the tree).

Verify with: python3 CHECK (exit 0 = pass).