Add cost and slippage stress scenarios to validation reports

[initial-d]

Context: the Awesome AI Trading Research full-text review included this caveat:

Linear transaction cost model may be optimistic for larger portfolios.

Goal: add a documented stress-test layer around transaction costs and slippage, so public validation reports can show how fragile a result is under less optimistic execution assumptions.

Suggested scope:

• keep the current linear cost model as the default baseline;
• add optional stress scenarios, such as higher bps, turnover-linked cost, liquidity haircut, or capped participation assumptions;
• report net metrics side by side with clear labels;
• document which assumptions are simple research stress tests rather than live execution models;
• add tests for cost accounting units and monotonicity where applicable.

Useful entry points:

• docs/backtest_assumptions.md
• docs/backtest_cost_drag_story.md
• docs/validation_dashboard.md
• src/mlquant/backtest/engine.py
• src/mlquant/backtest/metrics.py
• review comment: Paper suggestion: Machine Learning Enhanced Multi-Factor Quantitative Trading ohselab/awesome-ai-trading-research#1 (comment)

Acceptance criteria:

• A validation report can include at least one non-default cost/slippage stress scenario.
• Documentation keeps the model boundary explicit.
• Tests prevent unit confusion between daily, cumulative, and annualized cost drag.