v0.9.0

Changed

• Redesigned web UI: sidebar navigation, light/dark/auto theme, an Overview page (totals, daily throughput, workers,
  queues, running jobs), per-stream consumer tables and batch cards; fonts and assets are bundled, nothing is loaded
  from a CDN. The Actions placeholder page is gone and the processes table moved to /processes
• The web UI refreshes after each response instead of on a fixed clock, backs off while refreshes fail, and shares
  identical refreshes between viewers for COSMO_WEB_CACHE_TTL seconds (default 2, 0 turns it off). While NATS is
  unreachable it answers 503 at once instead of waiting on NATS, which kept threads of a host app's Puma busy
• The NATS client keeps reconnecting through an outage instead of closing for good after nats-pure's 10 attempts
  (about 20s), which left workers and the web UI disconnected until restarted; NATS_MAX_RECONNECT_ATTEMPTS overrides it
• API::Cron::Entry#schedule_headers is split into #schedule, the schedule: option of nats-pure's publish, and
  #headers, the dispatch headers; Cosmo uses nats-pure 2.7's header constants and ttl:/stream: publish options
• API::Stats::Metrics#daily sums every job class when called without one; API::Stream#consumers lists consumers
• Requires nats-pure 2.7, which ships the concurrent fetch fixes and struct fields Cosmo used to patch in, so the
  patches are gone. nats-pure 2.7 now sends stream and KV settings 2.5 dropped: see its changelog if your app passes any