Visual identity from real engine frames, lorenz example, PROOF_SHOT capture, Pages site

[Mattbusel]

Gives the repo a visual identity built from frames the engine renders itself, plus a GitHub Pages site and a small capture feature that made it possible without opening a visible window.

Before and after

Before: the README opened on dark screenshots of the convergence demo and two 4 MB GIFs.

After: the banner is two real frames: the sky example at 17:50 (sun +1.8 deg) and the new lorenz example.

What changed

• PROOF_SHOT capture (src/capture.rs). Any program on the engine can write its frames to BMP, read back off the GPU after post-processing, via env vars: PROOF_SHOT, PROOF_SHOT_AT, PROOF_SHOT_COUNT, PROOF_SHOT_EVERY, PROOF_SHOT_KEEP, PROOF_FIXED_DT, PROOF_HIDDEN (invisible, unfocused window), PROOF_WINDOW. Wired into run, run_with_overlay and run_ui. No public API changes; three unit tests.
• examples/lorenz.rs. 40,000 points advanced by math::attractors::rk4_step, coloured by speed, drawn into the HDR world pass.
• README. Banner, quick start first, figures captured today from sky, lorenz and convergence, a hand-drawn pipeline SVG, reference sections collapsed into <details>, status corrected to 0.2. All previous accurate content kept.
• docs/: a Pages site with the sky clip in the hero, a "drag the sun" scrubber over 40 real frames (times and elevations match what the demo prints), the Lorenz clip, quick start with copy buttons, features, demos, capture docs. No trackers, no external fonts.
• .github/social-preview.png (1280x640) to upload in Settings.
• Root screenshots with timestamp names removed; the convergence GIF moved to assets/. docs/ and *.jpg excluded from the published crate.

Found while capturing

galaxy, math_rain and strange_attractors render a blown-out first second and then an empty scene (captured hidden on Windows, RTX 4070; convergence and supernova on the same path look right). The README now says so instead of advertising them; the site lists only demos that were seen working or were already documented as needing a strong GPU.