v5.25.1

Fixed

• Hindsight records through an agent outage. A failed scrape exited
  Hindsight with status 1, so restarting the agent (a package upgrade, for
  example) discarded the buffer with no capture and left its staging
  directory behind; with Restart=on-failure each agent restart added
  another. A failed scrape is now a tick with no row: Hindsight keeps its
  buffer, keeps sealing and evicting, and tries again on the next tick. A
  scrape that gets no response within three intervals or 5 s, whichever is
  longer, returns an error status or does not decode counts as failed. The
  scrape had no timeout, so a hung agent also stalled the loop and stops. A
  scrape slower than the interval but inside the timeout is still recorded,
  as before. Hindsight logs a warning when the first scrape fails, a line
  when one succeeds again, and a warning when the agent's producer epoch
  changes (a restart between two scrapes fails neither). /status gains
  failed_scrapes and scrapes_failing_since. An agent that cannot be
  reached at startup still exits; the probe now runs before the staging
  directory is created, so that exit no longer leaves an empty one in
  buffer_dir.