Input: a line guideline G, then a CSV date,district,pm25.
Build: line = line chart (x = date, y = pm25, colour = district); rule = a dashed horizontal rule at G; line + rule; line faceted by column="district:N"; and (line + rule) | bar chart of mean pm25 per district.
For each composite print its top-level structure from to_dict(): line: single with 1 part(s), line + rule: layer with 2 part(s), facet: facet with 1 part(s) -> {'column': 'district'}, line+rule | bars: hconcat with 2 part(s) (the first of layer, facet, hconcat, vconcat, concat, repeat present; number of parts = length of that list, 1 for facet/repeat/single). Finally print days over guideline: Lucheng 4 (count of rows with pm25 > G per district, alphabetical, or none).