feat(analysis): name the sectors in the N/D comparison CSV, and give it a sortable magnitude

[WesIngwersen]

The n_d_nowcast_<year>_vs_v03.csv carried bare sector codes and percent differences only, which made the obvious question — which sectors moved most — awkward to answer.

Sorting on the percent column answers a different question. It puts near-zero sectors on top, because a small absolute move on a small base is a large percent. The magnitude of the move is what "moved most" means, and there was no column for it.

What is added

column
sector_name	Cornerstone commodity description
N_change / D_change	level change, nowcast − v0.3
N_abs_change / D_abs_change	its magnitude — the sort key

⚠️ Names come from the Cornerstone taxonomy, not BEA 2017. The model runs on the 405-sector Cornerstone axis and cornerstone.commodities.COMMODITY_DESC has exactly those 405; the BEA 2017 dictionary is a different 408 and would leave the Cornerstone-specific codes blank. The script counts unnamed sectors and prints a note rather than silently writing empty cells — on the current build it reports none, 405 of 405 named.

What it shows

Top movers by N_abs_change on the current build — chemicals, all down hard:

sector	name	N v0.3	N nowcast	change
327310	Cement	4.330	4.762	+10.0%
325130	Synthetic dyes and pigments	1.158	0.734	−36.6%
33131B	Processed aluminum products	0.940	0.547	−41.9%
111900	Tobacco, cotton, sugarcane and other crops	2.048	1.664	−18.8%
325320	Pesticides	0.981	0.612	−37.7%
325520	Adhesives	0.790	0.487	−38.4%
562213	Solid waste combustors and incinerators	5.144	5.439	+5.7%
331313	Primary aluminum	1.083	0.790	−27.1%

On D_abs_change the same chemicals lead and fall further still — adhesives −70%, pesticides −62%, dyes −49% — so these are direct-intensity moves rather than something arriving through the supply chain.

Testing

Analysis script, no unit tests per repo convention.

uv run python -m bedrock.analysis.nowcasting.results.nowcast_vs_v03_hists
    # 405 rows, 405 named, no unnamed-sector note
uv run black --check . ; uv run ruff check .    # clean

🤖 Generated with Claude Code

[bl-young]

see comment on #877 about the shifting the comparison to v0.3.2 instead of v0.3.1 (that change is really outside both of these PRs)