Audit ccusage-daily-kpi.timer + claude-usage-watch.timer resource cost vs value

claude-usage-watch.timer confirmed working (real pg data every ~15min) but costs ~13s CPU + up to ~1GB peak memory PER RUN, every 15min all day. Evaluate whether that cadence/cost is justified vs e.g. 30-60min, and whether ccusage-daily-kpi's daily digest output is actually read by anyone (Elazar report - MSG-54/KPI-39 in current backlog suggest attribution problems in this family already).