fix(backtest): correct CN drawdown Sharpe and Calmar metrics

[Pigbibi]

Scope

• F2 / RA-03 / RA-06: include initial NAV in drawdown, use annualized arithmetic mean for zero-rate Sharpe, and preserve negative CAGR in Calmar.
• Only three production formula lines and two existing test files; no dependency, lockfile, catalog, promotion authority, data, or historical artifact changes.

Validation

• Fresh base: 7fb104e.
• Isolated environment: uv sync --locked --offline --extra test; pinned QPK c812ed70f83d61bdf1816fa5ca112b0f6976c6b6; uv pip check passed.
• Existing targeted baseline: 9 passed. New regression RED: 5 failures. GREEN: 18 passed, 4 subtests.
• Hand-calculated oracles: MDD [-10%] -> -10%, [-10%,-10%] -> -19%, [+10%,-20%] -> -20%; arithmetic Sharpe [10%,-10%] -> 0 and [1%,2%,-1%] -> 8.485281374238571. Calmar negative/positive/zero and undefined zero-drawdown cases covered.
• Full offline suite: 102 passed, 4 subtests, 1 pre-existing checkout-path failure in test_package_root_exports_use_local_source_tree (expects CnEquityStrategies/src without the worktree directory). The test and package init are unchanged from main; not bypassed or modified.
• Full Ruff, diff, scope, compile, and added-line credential-pattern checks passed.

No provider, real replay, deployment, account operation, or historical evidence recomputation.