system-banner.tsx: downgrade transient connect-timeout catch error->warn (stops non-actionable real-time pager)

applog fired a real-time ERROR pager on systemBannerQueryFailed (timeout exceeded when trying to connect, 2026-07-04T17:32:51Z). db-pluto-cc CHARACTERIZED = lone transient Supabase pooler blip, NOT pool exhaustion, NOT the PLUTO-471 class. Facts: systemBannerQueryFailed = 6 rows all-time; the 17:32 event = a 40ms burst of 5 concurrent root-layout banner renders catching one instantaneous connect stall; ZERO other action types errored in the +/-3min window (real pool exhaustion would have connect-timed /informes//alumnos//practicas too); Supabase 18/60 conns, recovered instantly. SystemBanner runs in the ROOT LAYOUT with noStore() (always dynamic) = highest-frequency + first DB touch per render = an involuntary canary that catches any momentary blip first. USER IMPACT: none - src/components/system-banner.tsx already degrades gracefully (catch returns null, renders no banner, never crashes). The ONLY effect is a level=error log line that fires the real-time pager on a non-actionable transient. FIX (coder-lane, NOT db, NOT Class-A): downgrade THAT ONE catch in system-banner.tsx from error->warn. logCaughtError hardcodes error; use the logWarn helper (added PLUTO-461) or a level arg. Keep category/action + full error detail; ONLY the level changes. Rationale: decorative + fail-soft + result-discarded-on-error = non-actionable at error tier. Signal preserved: any pooler blip severe enough to matter also hits a user-facing query and pages via serverError (the 2026-07-02 /informes serverError row proves it). warn keeps it in the 30min digest for trend-watching without paging. No DB change, no migration. Standard PTD via audit-pluto-cc.