Add subplot / multi-panel layout (`Figure::subplot()`)

[MoonFlowww]

Summary

Add support for arranging multiple plots in a grid layout within a single Figure, matching matplotlib's plt.subplot() / fig.add_subplot() workflow.

Motivation

Performance comparisons routinely need side-by-side panels: throughput vs. latency, before vs. after, different input sizes. Without subplots, users must create separate PPM files and stitch them manually. This is a significant quality-of-life gap vs. matplotlib.

Proposed API

sepia::plot2d::Figure fig(1200, 800);

auto ax1 = fig.subplot(2, 2, 0);   // 2 rows, 2 cols, panel index 0 (top-left)
auto ax2 = fig.subplot(2, 2, 1);   // top-right
auto ax3 = fig.subplot(2, 2, 2);   // bottom-left
auto ax4 = fig.subplot(2, 2, 3);   // bottom-right

ax1.plot(x1, y1, n).label("throughput");
ax2.plot(x2, y2, n).label("latency");
// ...

fig.render();
fig.save_ppm("comparison.ppm");

Each Axes object returned by subplot() has the same API as Figure today (plot, scatter, bar, set_xlabel, etc.) but renders into a sub-region of the parent canvas.

Implementation Notes

• subplot(rows, cols, idx) computes a BBox (pixel rect) for that cell and stores it
• Returns an Axes handle that holds a reference to the parent Canvas and a CoordTransform scoped to that cell's pixel rect
• Each Axes renders its own grid, axis, and legend within its BBox
• Layout: divide Figure pixel area into a uniform grid, apply shared margins + inter-panel spacing parameter
• Figure::render() iterates all Axes and calls axes.render_into(canvas)
• Shared X/Y axis labels: subplot(2,1,0, share_x=true) suppresses X tick labels on the top panel

Acceptance Criteria

• subplot(rows, cols, idx) allocates correct pixel regions
• Each panel has independent data, axis, and legend
• Panels do not overdraw each other
• share_x / share_y suppresses redundant axis labels
• Example: examples/subplots.cpp with a 2×2 performance comparison grid