fix(kernel): pass cursor row limit to kernel

[vuanhphung]

Forward cursor row_limit to the kernel statement and remove the temporary connector-side stream truncation. This uses databricks-sql-kernel#269 and advances KERNEL_REV to the merged commit.

Validation: python -m pytest tests/unit -m 'not realkernel' (895 passed, 4 skipped). Live kernel coverage exercises None, 0, 1, and 5.

PECOBLR-4127

[peco-review-bot]

Verdict: 1 Medium

Looks good overall — the row_limit forwarding is wired correctly and the _drain() rewrite preserves buffer/offset/row-index accounting. One medium concern: the tests silently dropped the row_limit=0 boundary case, which the removed connector code treated as a real zero-row limit, leaving a possible behavior change (0 → unlimited, matching SEA) uncovered.

[peco-review-bot]

Verdict: 1 Low

Looks good — forwarding row_limit to the kernel and removing the connector-side stream truncation is coherent: set_row_limit is applied before both the sync and async execute branches, the _drain rewrite accounts for _buffer_offset/_buffered_count/_next_row_index consistently, and unit + e2e tests are updated (including the new None case). One low-severity note: negative row_limit was previously normalized to "unlimited" and that behavior is now dropped and untested.