Add histogram support (`figure.hist()`)

[MoonFlowww]

Summary

Add figure.hist() for plotting frequency distributions from raw sample data, matching matplotlib's plt.hist().

Motivation

Histograms are the standard first step when analysing latency distributions from benchmark runs. The user passes raw measurement samples and the library handles binning automatically.

Proposed API

std::vector<double> latencies = /* ... benchmark samples ... */;

figure.hist(latencies.data(), latencies.size())
      .bins(50)                              // number of bins (default: auto)
      .range(0.0, 200.0)                     // optional clamp range
      .color(sepia::Color{100, 200, 120})
      .alpha(0.75)
      .label("latency distribution (ns)");

Auto bin count should default to Sturges' rule: ceil(log2(n) + 1).

Implementation Notes

• Add HistEntry (or tag PlotEntry with a kHistogram type)
• Binning logic: compute [min, max] of data (or use user range), divide into k equal-width bins, count samples per bin — O(n) pass
• Rendering: call canvas.fill_rect() + canvas.draw_rect() (outline) for each bin, similar to bar()
• Normalized mode (density=true): divide counts by n * bin_width to get probability density
• Stacked histograms: multiple hist() calls with alpha < 1 already work via existing alpha-blending

Acceptance Criteria

• Auto bin count via Sturges' rule
• User-supplied bin count
• Optional range() clamp
• Optional density(true) normalization
• Multiple overlapping histograms with alpha blending work
• Example: examples/histogram.cpp with synthetic latency samples