fix: preserve gaps in multi-strategy reports

[JohnDReynolds]

Summary

Fix report preprocessing so a missing observation in one strategy does not remove or replace valid observations from another strategy.

Previously, report-level row deletion could discard an entire date when any strategy column contained a gap, while zero-filling could treat missing observations as real zero returns. This produced incorrect cumulative returns, volatility, and date metadata in multi-strategy reports.

Behavior

Each strategy is now evaluated from its own observed returns instead of a row-wise intersection of every strategy column.

The change:

• preserves per-column gaps through HTML, basic, full, metrics, and plot report paths;
• stops replacing missing report observations with zero returns;
• forwards match_dates consistently from basic() and full() to plots();
• reports a Series start date from its first observed value; and
• operates on copies where report normalization modifies index or label metadata, preserving caller-owned inputs.

Complete-data and single-Series report behavior remains unchanged.

Regression evidence

• Fail before: In the retained six-day regression, Strategy A's observed returns are 1%, 50%, -2%, 3%, and 1%. They independently compound to (1.01 × 1.50 × 0.98 × 1.03 × 1.01) − 1 = 0.54453341, displayed as 0.54. Accepted report preprocessing instead displayed 0.02 when Strategy A appeared beside another strategy with gaps. Zero-filling also reduced annualized volatility from the direct result of 3.51 to 3.21, and a leading gap caused the report to show 2024-01-01 instead of the first observation on 2024-01-06.
• Independent expectation: For report-preprocessing behavior, each column in a combined strategy table must match that same strategy evaluated alone. Cumulative return and volatility are checked independently with stats.comp() and stats.volatility().
• Pass after: 10 focused cases pass across basic and full modes, both match_dates settings, gapped benchmarks, leading/interior/all-strategy gaps, caller-input preservation, report delegation, and structurally parsed HTML output. Full-table comparisons exclude the existing Smart Sharpe and Smart Sortino rows because their metric-specific cross-column behavior is not changed here.

Verification

• 10 focused regression tests passed.
• 145 report, regression, and plot tests passed.
• 235 full-suite tests passed with 72 inherited dependency deprecation warnings.
• Changed-file and full Ruff lint passed.
• Ruff formatting and git diff --check passed.
• Wrapper diagnostics passed.
• Generated HTML output was parsed structurally.
• No public signature, template, dependency, documentation, or build behavior changed.

Scope

Changed files:

• quantstats/reports.py
• tests/test_reports.py
• tests/test_regressions.py

Out of scope:

• Smart Sharpe and Smart Sortino metric-level penalty isolation
• Nullable extension-dtype handling in stats.ulcer_index()
• Redesigning match_dates
• Unrelated report formatting or cleanup