v1.0.0-rc.10

Fixed

• The plugin process no longer hangs forever when Tabularis quits while an
  RPC is in flight. After stdin EOF (the orphan path — the host is gone),
  in-flight RPCs now drain for at most 5s (SHUTDOWN_GRACE), then any
  worker still blocked on an unbounded await is force-cancelled via
  handle.abort(), which drops the held deadpool::Object and closes the
  DB socket server-side. tokio-postgres's connect_timeout bounds only
  the socket-level connect, not the startup handshake, so it could not
  bound the pooler-accepts-but-never-responds case — only this shutdown
  deadline covers every hang mode (connect, handshake, query, pool wait)
  uniformly (#156, #155).

Added

• pooler_mode field in ConnectionParams for PgBouncer transaction-mode
  pooler awareness. When the host declares "transaction" (companion:
  tabularis#1126), the plugin skips its own prepare() call for
  column discovery on empty result sets (eliminating the plugin's
  named-statement trigger) and rewrites SQLSTATE 26000/42P05 errors
  (from tokio-postgres's internal typeinfo_statement fired by non-builtin
  column types — custom enums, composites, extension types) with actionable
  guidance pointing to max_prepared_statements > 0 in pgbouncer.ini.
  Backward-compatible: pooler_mode defaults to None — no behavior
  change until the host sends the field (#158, #157, tabularis#1097).