Avoid unnecessary sampling extent scans

Copilot review identified two edge cases in the sampling count path: missing base-axis dimensions should not collapse the rate to zero, and filtered data whose extent already lies inside the axis window does not need another full count pass.

This keeps none/empty mode on the visible-window scan path, while allowing filter mode to use the already filtered data count when the extent proves it is safe.

Constraint: dataZoom filterMode none/empty keep raw data count even when the axis window is narrow
Rejected: Use count whenever count differs from raw count | unsafe when the current data extent is not fully inside the base-axis scale
Confidence: high
Scope-risk: narrow
Directive: Keep countDataInAxisExtent as the fallback for unfiltered and extent-ambiguous dataZoom modes
Tested: npm run checktype -- --pretty false; npm run lint -- --quiet src/processor/dataSample.ts; git diff --check -- src/processor/dataSample.ts; Chrome headless HTML regression screenshot
Not-tested: Full visual regression suite