docs(book): the lighting page stops contradicting itself about shadows

[lobinuxsoft]

788 lines, and the parts written before #777/#778/#780 were arguing with the parts written after.

• 🔴 It said a punctual light needs a cube map "and neither exists yet" — then documented the cube map four hundred lines later. Both statements were on the page. Now: cascades are for the first directional light, spots cast through their own map, points through a cube, both cast_shadows: true by default and both limited to four.
• Clustering is no longer written in the future tense ("when clustering lands its input is…").
• Exposure and ambient are no longer "a control not in your hands" — feat(render,ecs): render settings as a project asset, with defaults #744 made them a .rendersettings asset. Documents why an asset and not a panel: bespoke settings panels do not get built, and this setting had none for as long as it existed.
• ❌ New section: what does not work — cutting the froxel list by count (❌ Sample the tile's lights instead of walking all of them — refuted #826). RIS over a cell's lights is the obvious next idea, it was built and removed, and the reason is structural: it breaks the grid's continuity property and the froxels become visible. With the histogram, and the LDS hypothesis that died with it.
• The tonemap is inline on the fragment path and its own pass on the compute one.

Third page of the pass over the book. Docs only.