docs: clarify benchmark weighting semantics

[dajinzhu]

Summary

Related to #2315.

• Clarify that a list benchmark is calculated as an equal-weighted average of the supplied instruments.
• Clarify that a string benchmark uses the return series of one benchmark instrument.
• Document the weighting effect when an equal-weighted strategy is compared with a cap-weighted index.
• Add a regression test covering the list-benchmark equal-weighting semantics.

This pull request intentionally does not change Qlib's default SH000300 benchmark or historical backtest behavior. A dynamic point-in-time equal-weight benchmark would require a separate design decision.

Testing

• git diff --cached --check
• python3 -m py_compile qlib/backtest/report.py tests/backtest/test_benchmark_weighting.py
• PYTHONPATH=.../.openclaw/tmp/qlib .../.openclaw/tmp/qlib-venv/bin/python -m pytest -q tests/backtest/test_benchmark_weighting.py

Targeted test result: 1 passed.

[dajinzhu]

@microsoft-github-policy-service agree