fix: anchor drawdowns at high-water marks

[stefan-jansen]

What changed

• Anchor each reported drawdown period at its actual high-water mark.
• Preserve that peak while losses remain below the reporting threshold.
• Add a regression test covering delayed threshold crossing.
• Bump the package version to 0.1.0b25.

Why

The prior implementation set peak_date to the observation immediately before a drawdown crossed the configured threshold. For gradual declines, this shortened the reported drawdown and assigned the wrong peak date. The corrected scan continuously tracks the latest high-water mark outside a drawdown.

User impact

Drawdown tables and downstream reports now show the true peak date and full peak-to-recovery duration. Maximum drawdown depth is unchanged.

Validation

• Reproduced b24 behavior: Jan 2 peak, 2-day duration
• Verified b25 behavior: Jan 1 peak, 3-day duration
• uv run pytest tests/test_evaluation/test_portfolio_analysis.py -q - 70 passed
• uv run pytest tests/ -q -n auto --timeout 120 - 5,287 passed, 21 skipped
• uv run ruff check src/ tests/test_evaluation/test_portfolio_analysis.py
• uv run ruff format --check src/ tests/test_evaluation/test_portfolio_analysis.py
• uv run ty check
• pre-commit run --all-files
• Built wheel and sdist; installed-wheel smoke test passed for version and corrected drawdown output

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.