Avoid redundant full-extent sampling scans

Copilot review pointed out that the previous fast path still scanned when the processed data count matched the raw count, even if the base-axis scale already contained the data extent. The extent check alone is enough to decide whether a visible-window count is needed.

This preserves the scan for dataZoom none/empty narrow windows, while avoiding it for the fully visible or already-filtered cases whose data extent is inside the current axis extent.

Constraint: dataZoom none/empty can keep raw count while the visible axis window is narrow
Rejected: Keep raw-count equality as a scan trigger | it regresses the fully visible common path
Confidence: high
Scope-risk: narrow
Directive: Use the base-axis extent relationship, not raw-count equality, to decide whether the count scan is necessary
Tested: git diff --check -- src/processor/dataSample.ts; npm run lint -- --quiet src/processor/dataSample.ts; npm run checktype -- --pretty false; Chrome headless HTML regression screenshot
Not-tested: Full visual regression suite

Author: susiwen8

src/processor/dataSample.ts

@@ -131,8 +131,7 @@ export default function dataSample(seriesType: string): StageHandler {
                 const baseDim = data.mapDimension(baseAxis.dim);
                 let dataCount = count;
                 if (baseDim != null) {
-                    const rawCount = seriesModel.getRawData().count();
-                    if (count >= rawCount || !dataExtentInAxisExtent(data, baseAxis, baseDim)) {
+                    if (!dataExtentInAxisExtent(data, baseAxis, baseDim)) {
                         dataCount = countDataInAxisExtent(data, baseAxis, baseDim);
                     }
                 }