Clarify Best/Worst Trade [%] definition and add regression tests for #810 - #875

Test evidence

Ran the new regression test plus all trade-related tests in tests/test_portfolio.py:

$ python -m pytest tests/test_portfolio.py::test_best_trade_pct_definition -v
tests/test_portfolio.py::test_best_trade_pct_definition PASSED [100%]
1 passed in 63.35s

$ python -m pytest tests/test_portfolio.py -k "trade" -q
4 passed, 178 deselected in 24.49s

Environment: fresh venv, Python 3.12, numpy/pandas/scipy/plotly/numba installed from pyproject.toml dependencies.

To clarify the semantics (issue #810): Best Trade [%] is the maximum of individual trade returns, where return = PnL / trade entry value — not PnL / portfolio init_cash. This PR only clarifies the docs and adds regression tests; no calculation logic was changed.

The definition in the new comment and docstrings is right for the zero-fee case, but the tests never leave it, and the docstring "return = trade PnL / trade entry value" can mislead once fees exist. In get_trade_stats_nb (portfolio/nb.py ~L5711):

entry_val = size * entry_price
...
pnl = val_diff - entry_fees - exit_fees
ret = pnl / entry_val

So the numerator is net of both fees and the denominator is the gross size * entry_price, with entry_fees not added to it. Two consequences worth pinning in a test: a short trade's return is (entry - exit)/entry - (entry_fees + exit_fees)/(size*entry), not the plain price ratio asserted in test_short_trade_return_math; and a trade can rank first by PnL, second by return, and flip again after fees if the fee rate differs by size. Both tests set fees to 0 (test_best_trade_pct_definition relies on the default), so a change to the denominator, say to include entry fees, would pass.

I'd add one case with fees=0.001 and assert Return == (pnl)/(size*entry_price) against hand-computed numbers, and say "net of fees, relative to gross entry notional" in the comment.

Smaller one: the identity stats()["Best Trade [%]"] == trades.closed.returns.max()*100 holds because the default settings use closed trades; with incl_open=True it will not. The test has no open trade, so it can't tell the difference.

Read from the diff and nb.py on master, not run.

Thanks for the helpful review, @arhancanli! Both points are addressed in f5a705e:

• Added a fees=0.001 case with hand-computed PnL, verifying Return == pnl / (size * entry_price). Updated the comment and docstrings to say "net of fees, relative to gross entry notional".
• Added tests for both incl_open=True and incl_open=False, covering Best/Worst Trade [%] with open trades.

Relevant tests: 7 passed, 178 deselected.