v0.2.0

Added

• jev_batch: the same questions put to up to 100 items in one tool call, one request
  per item. Answers and errors are keyed by the caller's item ids, and one failed item
  does not sink the rest.
• Batches are paced: OPENJEV_BATCH_CONCURRENCY requests in flight (default 4), and a
  429 pauses new requests for the interval the service asked for and halves the
  concurrency. A batch stops early, keeping its answers and listing the unsent items, on a
  rejected key, a rate limit it cannot wait out, or three failures in a row. It is
  bounded by the single-call worst case (~120 s by default), so existing client timeouts
  cover it, and it reports progress when the client asks for it.
• examples/: real DeepSeek Harness transcripts of an electric co-op's help desk, one
  member complaint through jev_ask and a morning's queue through jev_batch, with the
  tasks and the overlay that recorded them.

Changed

• jev_ask is titled "Ask Jev (several questions)", so it is not mistaken for jev_batch.
• The jev_ask and state descriptions say that a state is judged as one whole and
  point a list of records to jev_batch. A recorded agent had packed a whole ticket
  queue into one state and got one blended answer.
• Releases publish themselves. npm run release -- <version> sets the version in every
  manifest and turns the Unreleased entries into its notes; merging that to main
  publishes it to npm and the MCP Registry and creates the tag and GitHub release.
  The version check now covers package-lock.json too.