The consumption export keeps changing filename (每日数据 -> Week39...),
which broke the glob each time. build_data.py now treats any workbook
containing a 平台消耗统计 sheet as a data source and orders them by the
latest date they contain, so the fresher file wins on overlapping days.
2026Week39平台消耗.xlsx covers 8/31-9/27 and fully supersedes the 2609
file, which is removed; 2608 stays for the earlier August dates.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
