v0.1.2

Fixed

• The agent skill's setup: a STOP after the server or container started now
  stops it, and when the data was fresh it leaves it fresh — the database this
  start made is removed, never one that was there before or one another
  server holds — so the next try neither meets a container named … is there
  already nor mistakes the empty database for an earlier install. An earlier
  install's data is no longer started by the skill: the human starts it and
  gives its URL and key, or the project gets a new data directory. Keys are
  read from this start's log lines only.
• The agent skill's upgrade: the release fetched to plan with goes to a new
  directory of its own per upgrade (~/.cache/tracepad/tmp.XXXXXX, printed as
  BRIDGE:), so two sessions upgrading at once no longer take each other's
  binary, and a failed download stops there instead of running an older one.
  After the binary is replaced, the skill plans again with the new one and
  shows its advice.
• tracepad upgrade --plan: a tracepad first on PATH that a package
  manager owns through a link is given its manager's command, never the
  install script's line; that note now appears when a plan is refused too, and
  a binary that does not say its version says why. A local build stamped with
  an unreleased version is refused by the plan, as by the run.

Changed

• tracepad upgrade --plan says, in one line, when the skill's copies are
  newer than a server or container that is yours to upgrade. It does not
  change the exit status.
• traces show carries no scores, by design; the agent skill and
  docs/cli.md point at scores ls --trace.

Every commit since v0.1.1: v0.1.1...v0.1.2