fix(portfolio): handle NaN trade plot markers

[JohnDReynolds]

Fixes #716.

Summary

Prevent non-finite open-trade returns from producing invalid Plotly marker metadata in trade-PnL plots.

Finite returns retain their existing relative marker scaling. Trades with non-finite returns receive the lower marker-size and opacity bounds, while their underlying PnL and return values remain unchanged.

Problem

A portfolio with an open trade whose final valuation is unknown can produce a non-finite trade return:

import numpy as np
import pandas as pd
import vectorbt as vbt

close = pd.Series([1.0, np.nan])
pf = vbt.Portfolio.from_orders(
    close,
    [1.0, 0.0],
    ffill_val_price=False,
)
pf.plot_trade_pnl()

On master, this raises a Plotly ValueError because NaN is passed as marker opacity.

Trades.plot_pnl() previously scaled all absolute returns together. A single non-finite return therefore contaminated the minimum and maximum used to calculate marker sizes and opacities.

Change

• Identify finite absolute returns before scaling.
• Initialize marker sizes and opacities with their valid lower bounds.
• Apply the existing relative rescaling functions only to finite returns.
• Preserve the original PnL, return, grouping, and column-selection behavior.
• Add regression coverage for:
  • the exact all-non-finite issue reproducer;
  • mixed finite and non-finite trades;
  • direct Trades.plot_pnl() usage;
  • delegated Portfolio.plot_trade_pnl() usage;
  • finite marker metadata across all emitted traces.

Independent numerical check

For two long trades moving from 1 -> 2 and 1 -> 3, the independently calculated returns are [1.0, 2.0].

They continue to map to:

• marker sizes: [7.0, 14.0];
• marker opacities: [0.75, 0.9].

The non-finite open trade receives the lower bounds:

• marker size: 7.0;
• marker opacity: 0.75.

Grouped direct, delegated, and Portfolio.plot() orchestrated paths were also checked with physical-column selection.

Verification

• Focused regression: 1 passed, 84 deselected
• Complete tests/test_plotting.py: 85 passed
• Full Python suite: 941 passed, 100 skipped
• git diff --check: passed

The changed-file Black check reports that both files would be reformatted, matching the unchanged upstream baseline. No unrelated formatting changes are included.

Engine and scope implications

This change affects only Python/NumPy/Plotly presentation logic after trade records have been generated.

It does not change:

• trade accounting or simulation;
• pandas accessor behavior;
• global settings;
• broadcasting or grouping semantics;
• generic rescaling helpers;
• Numba kernels or engine dispatch;
• Rust code or engine parity;
• dependencies or packaging.