…V to batch 2

Connector half of the kernel batch-2 fixes (kernel PR #123). Bumps
KERNEL_REV to pick up the batch-2 kernel surface.

H4 — don't close the kernel Statement at sync execute-return:
execute_command's finally used to `stmt.close()` immediately after
`stmt.execute()` succeeded. For a large CloudFetch result with
paginated chunk links (all_fetched=false), the kernel fetches later
links lazily (get_result_chunks against the LIVE statement) during
consumption, so a premature CloseStatement broke those fetches. The
kernel now auto-closes the server statement when its result stream
drains (ExecutedStatement::next_batch end-of-stream), with the
executed-handle Drop as the backstop for partial/abandoned reads. So
the connector now flips close_stmt=False on a successful execute and
only closes on the error path (no executed handle was produced).

The other batch-2 fixes (cancelled-class -> OperationalError, U2M
refresh fail-fast, metadata statement close-on-drop, per-binding OAuth
client_id) are entirely kernel-side and need no connector code beyond
the KERNEL_REV bump.

Tests: unit (sync execute does-not-close on success / does-close on
failure) + e2e (large multi-chunk result drains without premature
close + cursor reuse; server cancel maps to OperationalError not
ProgrammingError). All e2e verified live against dogfood.

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

#123 picked up a follow-up commit fixing the wiremock cancelled-state
assertions (ErrorCode::Cancelled). Bump the placeholder pin so the
connector CI builds the corrected kernel. Still to be re-pinned to the
squash-merge SHA before #830 merges.

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

Kernel batch-2 (#123) is merged to kernel main (f4ee6fe, current main
HEAD). Re-pin from the orphaned branch HEAD (4f7fbe7) to the merged
SHA — reachable from main, no orphan-SHA risk. Verified against a
wheel built from f4ee6fe: connector unit (102) + kernel e2e (H4
large-result drain + reuse, server-cancel -> OperationalError, staging
fail-loud, diagnostic-info) all pass against the real merged kernel.

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

…atch2-fixes

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

Address review: 'H4' is internal audit shorthand, meaningless in the
public connector codebase. Reword the affected comment + two test
docstrings to describe the behavior directly (premature sync
CloseStatement broke lazy CloudFetch chunk-link fetches). No code change.

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