Keep normalized stack totals linear under dense area data

Copilot review identified that normalized stacked area calculation rescanned every
series for each point. The processor now precomputes per-stack totals by raw
index and by stack dimension, with sign buckets reused by each stack strategy.
A regression test locks the accepted all-series opt-in behavior.

Constraint: Do not modify dist; release build artifacts are generated during npm publish
Rejected: Keep per-point stack rescans | scales as series^2 * points for normalized stacks
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep stackNormalize denominators aligned with stackStrategy sign buckets when changing stack semantics
Tested: npx jest --config test/ut/jest.config.cjs --coverage=false test/ut/spec/series/lineStack.test.ts
Tested: npm run checktype
Tested: npx eslint src/processor/dataStack.ts src/chart/line/LineSeries.ts test/ut/spec/series/lineStack.test.ts
Tested: git diff --check src/processor/dataStack.ts test/ut/spec/series/lineStack.test.ts
Not-tested: Browser visual screenshot against a fresh source bundle; avoided regenerating dist per repo rule