v0.7.0

Added

• Processes page in the Web UI listing every live worker with its host, IP, command line, subscriptions,
  busy threads, and state, fed by a heartbeat into the cosmo_processes KV bucket
• Cosmo::Client names its NATS connection (cosmo-<program>-<host>-<pid>), so a process can be
  identified in nats server report connections / /connz without cross-referencing host IPs.
  Override with COSMO_CLIENT_NAME
• COSMO_JS_TIMEOUT and COSMO_CONNECT_TIMEOUT make the JetStream API and TCP connect timeouts
  configurable. Defaults are unchanged (5s / 2s, as in nats-pure); raise the former per-process
  where waiting beats failing -- the web UI on a busy server -- and leave it alone for workers
• COSMO_WEB_POLL_INTERVAL sets the web UI's htmx auto-refresh interval (default 5s, unchanged).
• Enqueued jobs can be removed one by one from the web UI: each row has a Remove button backed by
  DELETE /jobs/enqueued/<seq>?stream_name=<stream>, which drops that message from the stream
• kill -TTIN <pid> logs every thread's backtrace at WARN, tagged with the same tid as regular log
  lines, so a stuck worker can be diagnosed without restarting it
• kill -TSTP <pid> quiets a worker (no new jobs are fetched, in-flight work finishes) and kill -CONT <pid>
  resumes it; kill -USR1 <pid> quiets it and exits once in-flight work drains
• Jobs workers can pick their streams with --streams a,b / --stream a or COSMO_JOBS_STREAMS; unknown names
  abort before boot. --no-scheduler keeps a worker off scheduled dispatch, so it fires no delayed jobs or crons
• Cosmo::HTTPServer: -p/--http-port (or http.port) serves GET /health (200 when the engine runs and NATS
  is connected, 503 otherwise) and GET /ping (liveness)
• Cosmo::Batch.new(bid:) takes your own batch id instead of a generated one
• The busy jobs page is paginated, shows each job's delivery attempt, and can stop live polling

Changed

• Relicensed from LGPL-3.0 to MIT
• Dashboard stats live under API::Stats on top of base classes: API::Busy is now API::Stats::Busy,
  and API::Counter.instance/#with moved to API::Stats::Totals
• nats-pure is capped below 2.7, and Cosmo's nats-pure patches skip themselves where nats-pure already has the fix
• rack, rackup and webrick are no longer runtime dependencies: add rack to your Gemfile for the web UI,
  and rack, rackup and webrick for Cosmo::HTTPServer

Fixed

• The _cosmostats counters stream kept every increment forever (one message per processed job). It now keeps
  only the latest message per counter (max_msgs_per_subject: 1); cosmo --setup applies this to existing
  streams, trimming them without changing the totals
• A cron job's enqueued_at is the time NATS fired the schedule again, not the time the scheduler
  dispatched it on. The firing's timestamp is forwarded as X-Enqueued-At across the re-publish, so a
  job that derives anything from enqueued_at is unaffected by a backlogged or restarted worker
• Dead jobs record the exception class, message and backtrace, so the web UI's Error column is no longer empty.
  The dead jobs page shows the stream a job came from and keeps expanded errors open
• API::KV#size stopped at 25 keys, so the web UI showed at most 25 busy jobs
• The dashboard's Scheduled count included cron templates, which the scheduled jobs page doesn't list
• The scheduler dropped a job's headers when dispatching it from the scheduled stream
• Several idle threads polling the same stream advanced its fetch backoff once each per round, reaching the
  maximum backoff after a few rounds; it now advances once per round
• Message#nack in stream processors raised NoMethodError; it is now an alias of nats-pure's nak
• JSON parsing works with json 3