Build and prove the 2D World->image rendering path

[tigercosmos]

Build and prove the 2D World->image rendering path

Related to #754.

Intention

Make modmesh's 2D drawing API produce a deterministic image of a World that both the live pilot widget and an offscreen renderer share. This is the foundation for the AI-friendly drawing work: the corpus images an AI model reasons about must be the same pixels a user sees, by construction.

Today R2DWidget renders a World only inside paintEvent (private paintWorld, coupled to widget state). There is no shared render routine, no offscreen QImage path, and no pixel-level test. This milestone closes those gaps and adds the overlay and text representations the experiment's input arms need (A2 overlay, A3 basic state, A4 state+diagnostics; A1 is the raw rendered image).

The work is split into small PRs (each < 500 LOC). PR3 is the gate: nothing downstream is trusted until a known World renders to expected pixels.

PR list

• PR1 - Extract shared paint(QPainter&, World&, view) routine. Move the body of R2DWidget::paintWorld into a free function with explicit arguments. Pure refactor, pixels unchanged, existing tests stay green. pilot: Extract world render logic into RWorldRenderer2d (issue #754 PR1) #899
• PR2 - Offscreen QImage renderer + Python binding. Render a World to a QImage via the PR1 routine with no event loop, and expose a 2D image capture binding (the current capture is R3DWidget-only). Antialiasing toggle for deterministic corpus images. pilot: Add offscreen 2D world renderer and Python binding (issue #754 PR2) #900
• PR3 - Real pixel test (gate). Replace the misleading skipped test at tests/test_pilot.py:47. Build a known World, render via PR2, assert non-blank pixels and that known endpoints map to expected pixels under the view transform.
• PR4 - Non-diagnostic annotation overlay (arm A2). Toggleable overlay: shape ids, axis labels, world-coordinate ticks, per-shape bbox, highlight-by-id. Legibility only - no derived facts.
• PR5 - World.describe_state(level="basic") (arm A3). Serialize shapes and control points to structured text - only what is also visible in the image, just as text. Add World::describe_state for visible 2D geometry #902
• PR6 - describe_state(level="+diagnostics") (arm A4). PR5 plus derived facts: intersections (reusing the polygon_boolean sweep-line) and degeneracies. The "oracle-ish" arm, kept separate so its advantage is attributable to information, not legibility.