fix(kernel): server-sourced async state + cursor-state/metadata parity for use_kernel (#838)

fix(kernel): cursor-state tracking + metadata parity for use_kernel

Brings the use_kernel=True path closer to Thrift parity on the
cursor-state-tracking and metadata clusters from the API+CUJ gap
audit, surfaces DML rowcount, and makes async command state
server-sourced.

Async command state — server is the source of truth (no local state):
- get_query_state / get_execution_result now re-attach to the
  statement by id (Session.attach_async_statement, kernel side) and
  read state / await results from the server, instead of consulting a
  connector-side handle map. SEA keys GetStatementStatus/Result purely
  on the id, so this works whether or not the connector still holds the
  submitting handle — and get_execution_result is inherently re-callable
  (each call re-attaches), matching the Thrift backend's re-fetchable
  operation handle.
- Deleted _closed_commands (and its bounded-FIFO machinery
  _record_closed / _CLOSED_COMMANDS_MAX): the connector no longer
  invents a CLOSED-vs-SUCCEEDED answer. The server reports CLOSED
  natively (verified: GET after close returns 200 state=CLOSED until
  TTL, then 404 NOT_FOUND, which maps to a terminal state). This is the
  local-state inconsistency removed.
- _async_handles / _async_statements are retained but DEMOTED to a
  keep-alive-only registry: the submitting ExecutedAsyncStatement's Drop
  fires close_statement, which would kill the live async query, so we
  hold the handle until close_command / close_session. They are no
  longer consulted for state or results.

Cursor-state tracking (T7):
- Set cursor.active_command_id in the _make_result_set chokepoint so
  sync execute, async fetch, AND metadata all leave the cursor pointing
  at the command that produced the current result set (matches Thrift's
  unconditional set in _handle_execute_response).
- On a failed sync execute, publish the server-issued statement id
  (read from the canceller's inflight slot) onto active_command_id
  before re-raising. Best-effort; never masks the original failure.

Metadata:
- Normalize wildcard/blank catalog ('%' / '*' / '' / None) to None
  (all-catalogs) across get_schemas/get_tables/get_columns
  (_catalog_or_none), making the three symmetric (fixes columns).
- Normalize empty/whitespace-only pattern args to None (match-all) via
  _none_if_blank. % / * stay as real LIKE wildcards on patterns.
- Drop the connector-side table_types drain + client-side refilter in
  get_tables; the kernel filters table_types itself (case-insensitively
  as of the batch-3 kernel change). Removed the dead
  _drain_kernel_handle / _StaticArrowHandle helpers + unused imports.
- Fix the Cursor.columns() docstring: catalog_name=None is accepted on
  all backends.

DML rowcount:
- Surface the kernel's num_modified_rows as cursor.rowcount for DML
  instead of the hardcoded -1. None (SELECT / warehouses that don't
  report it) leaves rowcount at -1, matching Thrift; getattr-guarded.

Bumps KERNEL_REV to the batch-3 kernel commit (case-insensitive
table_types, HTTP error-envelope parsing, num_modified_rows getter,
Session::attach_async_statement).

Co-authored-by: Isaac

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>

Author: vikrantpuppala

KERNEL_REV

@@ -1 +1 @@
-f4ee6fec78aabce8c0ea9c1ff47fc11b8191d013
+7ffb30d533c08651ca707b8dd13894c9e01cb68e