Changes in version 0.0.1                        

  - Host a DuckDB workflow database over Quack or use an embedded
    connection.
  - Submit idempotent task IDs and execute named R handlers through
    leased claims.
  - Replay native DuckDB values, suspend with durable sleeps, and
    recover expired leases.
  - Preserve R value mappings and allow explicit serialization through
    BLOBs.
  - Reject list and data frame names that collide under DuckDB's ASCII
    case-insensitive comparison during value admission.
  - Document the driver and engine compatibility requirements for
    VARIANT storage.
  - Fence stale task writes and support retry budgets and cancellation.
  - Optionally run bounded expired-lease maintenance with a DuckDB C API
    v1.2 coordinator and an explicit start, status, and stop lifecycle.
    Development builds load only into databases opened with
    allow_unsigned_extensions = TRUE.
  - ca_work() retries statements aborted by known write conflicts, with
    bounded jittered backoff (conflict_retries), so competing workers no
    longer exit on routine contention.
  - ca_process() supervises an external command in an attempt directory,
    renewing the lease while it runs and killing it on lease loss,
    cancellation, or timeout.
  - ca_tasks() lists task metadata without payloads, and ca_result()
    reads one completed result.
  - ca_close() releases every owned resource even when one cleanup step
    fails, reporting failures as canard_close_error.
  - Stored values that cannot be restored raise canard_restore_error, a
    storage error, instead of being recorded as handler failures.
  - In-memory databases and Quack clients now receive the package's
    DuckDB configuration. Clients previously ran with DuckDB's defaults
    and silently downloaded the httpfs extension that Quack clients
    load; install httpfs explicitly alongside Quack (make quack does
    both).