BUG: Return 0 from Trade.pl_pct when entry_price is 0 or NaN

[huixin3]

Fixes #1318.

Trade.pl_pct divided by entry_price unconditionally, raising
ZeroDivisionError when it was 0 and returning NaN when it was NaN.
It now returns 0 in both cases.

Added a regression test test_trade_pl_pct_zero_entry_price
covering both inputs; verified it fails without the fix.

[arhancanli]

The guard fixes the ZeroDivisionError, but returning 0. changes what the stats see. _stats.py:71 builds ReturnPct from t.pl_pct, and _stats.py:91 takes returns = trades_df['ReturnPct'] for the trade-level numbers (win rate, average and best/worst trade, expectancy, SQN). A NaN pl_pct was skipped by pandas' mean/std, so a bad-entry trade dropped out of those; a 0. is counted as a real break-even trade, which pulls win rate down and shrinks the mean and SQN, and nothing in the output says a trade had an unusable entry price.

So for the NaN case the new behaviour is arguably worse than before: it turns "no information" into "a trade that returned exactly 0%". Returning float('nan') for both inputs (and keeping the guard so zero no longer raises) would keep the stats honest, and the test could assert np.isnan(trade.pl_pct) plus that stats['# Trades'] still counts it while Win Rate [%] is computed over the non-NaN trades. If you do want 0., a docstring line saying such trades count as break-even would at least make it a documented choice.

Smaller point: Trade.pl (line 663) still uses entry_price without a guard, so with NaN it returns NaN while pl_pct now says 0; the two disagree for the same trade.