[Bug]: Jinpan AP Window queries repeatedly exceed 7200-second execution limit

## Description

Multiple independent Window-function query shapes in the Jinpan AP workload reach the configured 7200-second limit on the dev multi-CN environment. This issue intentionally excludes `q1004_adv_window_frame_running`, which is already tracked in #26066.

## Environment

- Workflow: https://github.com/matrixorigin/mo-nightly-regression/actions/runs/33482459227
- Workflow branch: `codex/mo-ap-regression-dev-assets`
- Deployment: cn-dev `freetier-01`, multi-CN, OSS-loaded `jinpan_001` dataset
- SQL limit: `--sql-timeout 7200`
- Date: 2026-09-01 to 2026-09-04

## Reproduction

Run the Advanced Join phase against the `jinpan_001` dataset. The following Window-function cases each ran for the full 7200 seconds and were then terminated by the runner:

- `q1045_adv_window_frame_running`
- `q1127_adv_window_frame_running`
- `q1143_adv_inventory_ratio_win`
- `q1225_adv_inventory_ratio_win`
- `q1250_adv_window_frame_running`
- `q1348_adv_inventory_ratio_win`
- `q1430_adv_inventory_ratio_win`
- `q1824_adv_window_frame_running`
- `q1922_adv_inventory_ratio_win`
- `q1947_adv_window_frame_running`

## Actual behavior

Each listed query exceeded two hours of execution time. The runner emitted:

```text
Execution timeout after 7200s; skips remaining SQL in the current file and continues with the next file...
```

The completed Advanced Join shards contain repeated occurrences of this signature. No CN `OOMKilled`, exit 137, Go panic, or fatal error was found in the corresponding investigation windows.

## Expected behavior

Window queries over this AP dataset should complete within a bounded, operationally acceptable duration. A broad set of Window query shapes repeatedly consuming the entire two-hour test budget is a performance defect, even though the runner can cancel and continue.

## Scope and related issues

- `q1004_adv_window_frame_running` is tracked by #26066.
- This issue covers the remaining standard Window aggregate/order-by shapes.
- RANGE-frame cases are tracked separately in #28007.
- CN draining / connection loss is tracked separately in #27776.

## Evidence

- Advanced shard 1: https://github.com/matrixorigin/mo-nightly-regression/actions/runs/33482459227/job/99890422491
- Advanced shard 2: https://github.com/matrixorigin/mo-nightly-regression/actions/runs/33482459227/job/99890422678
- Case sources: `tools/mo-regression-test/cases/jinpan_001/join_adv/`

## Regression coverage

After a fix, add a bounded Jinpan/big-data performance regression for representative `window_frame_running` and `inventory_ratio_win` cases, with recorded execution duration and no two-hour timeout.