Add shared rolling-window engine for consistency and performance

[TDamiao]

Goal

Introduce a shared rolling-window engine to improve performance consistency and reduce duplicated math logic.

Why

Multiple indicators can reuse the same rolling primitives (mean/std/min/max/sum), reducing bug surface and warmup drift.

Implementation prompt

Create internal rolling core (e.g., src/core/rolling.ts) with reusable stateful + batch helpers:

• rolling sum/mean
• rolling std
• rolling min/max
• warmup-aware null behavior

Refactor selected indicators to use rolling core where it improves clarity/performance.

Scope

• Internal module first (no immediate public API required)
• Deterministic semantics matching current outputs
• Micro-benchmarks before/after for key indicators

Acceptance criteria

• No behavioral regression in golden tests
• Reduced duplicated rolling logic in overlap/volatility modules
• Benchmarks show equal or better performance on 10k/100k runs
• Dev notes document warmup/alignment invariants

Test plan

• Existing test suite + golden unchanged outputs
• New unit tests for rolling primitives and edge windows
• Benchmark comparison report attached to PR

[TDamiao]

Tracked by v0.4 master gate: #21.