[wow-miley]

Closes #153 (PHO-83). Builds on #175 (PHO-82, T12 · C).

What

• Modifier.lumosPointer(tiltDegrees = 0f, onInput) in :phosphor-lumos-compose, for a consumer drawing LumosCanvas directly. A Modifier.Node over SuspendingPointerInputModifierNode, so it is common to all six targets (onPointerEvent is skiko-only). It reads PointerEventPass.Final and never consumes.
  • Mouse and stylus: Enter/Move set the pointer, Exit sends NONE; a button press keeps the hover and pressed follows the button.
  • Touch: Press/Move set the pointer with pressed = true, Release sends NONE.
  • A change someone else consumed (a parent's scroll past touch slop) sends NONE, then the handler waits for every pointer to lift.
  • The handler's finally sends NONE. A new onInput on recomposition updates the node without restarting the handler.
  • Mapping: x = 2·px/w − 1, y = 1 − 2·py/h, then y = sy·cos t + √(max(0, 1 − x² − sy²))·sin t for a tilt.
• LumosCanvasFrame.pointer: Vector2? (@Transient, canvas px, Y down) for a material's light. PixelLattice.project(frame, pointer = frame.input?.pointer) carries it; PixelLattice.canvasPointer is the forward mapping, re-tilted and placed in the viewport, so the light lands on the voxel under the pointer.
• OrbStepper.setInput hands the pointer to the source and lays the latest one over every frame projected after it, live, still or recorded. NONE clears a stale pointer even on a live frame built before it left. A stepper never handed input leaves the frame's own pointer alone, so :phosphor-lumos-surface is unchanged. LumosDriver.setInput and a once-allocated onInput wrap it.
• LumosOrb carries the modifier on its Box with config.tiltDegrees, and a DisposableEffect(driver) tells the source NONE when the orb leaves composition or its source is replaced. No frame is published for a pointer move alone, so a paused or reduced-motion orb draws nothing extra.
• LumosInputPreview: three interactive THINKING orbs — alone, pointed at the centre, pointed upper right — each settled under its pointer.
• README: the Lumos section gains the pointer paragraph and a compiled PointedCanvas snippet, held to the quick start's rule by ReadmeLumosQuickStartTest.

Tests

./gradlew ktlintFormat :phosphor-lumos:jvmTest :phosphor-lumos-compose:jvmTest --offline is green locally; goldens unchanged.

• PixelLatticePointerTest (lumos commonTest): the frame's pointer in px, override and clear, tilt matches the front voxel and keeps the rim on the rim, past-the-rim on the equatorial plane, viewport, aspect.
• OrbStepperTest: setInput goes to the source and projects nothing by itself; the latest pointer lights live, still and recorded frames; NONE clears a stale live pointer; a stepper never handed input leaves the frame's own.
• LumosPointerMappingTest (compose commonTest): centre, corner, past the node, 30° tilt, and round trips through PixelLattice.canvasPointer tilted and untilted.
• LumosDriverTest: onInput reaches the source and lights the next frame in px.
• LumosOrbPointerTest (jvmTest, through SteppedScene.sendPointerEvent / sendTouch): hover at the centre (0, 0) and top-left (−1, 1), Exit → NONE; a mouse button follows pressed without ending the hover; touch press/release; an orb inside Column(Modifier.verticalScroll(state)) lets a touch drag scroll the column and reports NONE once the scroll takes over; removal reports NONE; a 30° tilt maps the front centre to y = sin 30°.
• LumosInputPreviewTest: the strip to build/lumos-previews/lumos-input-strip.png, and the desktop recording — a scripted pointer sweep driven through LumosOrb with real pointer events on desktop Skia, 90 frames at 30 fps — to build/lumos-previews/lumos-input-repel.gif, written through a small javax.imageio GifWriter (test-only).

Calls to review

• The parting barely shows on THINKING, as PHO-82/#152 T12 · C: interactive in VoxelFrameBuilder — a resting body under the pointer, VoxelFrame.input, GazingForm, LumosFrameSource.setInput, and a virtual pointer in LookDemo #175 flagged: the preset's skin jitter (±0.24) is larger than the repel's push (0.22). The strip test measures summed light in a 32 px box at the centre: 1594 alone vs 1569 pointed. The assertion holds, but the eye needs the marker ring in the GIF to find the pointer. The drawn-position repel follow-up from PHO-82/#152 T12 · C: interactive in VoxelFrameBuilder — a resting body under the pointer, VoxelFrame.input, GazingForm, LumosFrameSource.setInput, and a virtual pointer in LookDemo #175 is still Miley's call; this PR changes nothing in the builder.
• OrbStepper keeps the pointer rather than LumosDriver, so the rule lives framework-free with the other time rules, and the pointer wins over the frame's own once set. The ticket put it on the driver.
• Source replacement is covered by a DisposableEffect(driver) in LumosOrb rather than by restarting the pointer handler on a key: the handler keeps running with the new lambda, and the old source gets NONE.
• The scene collapses moves at an unchanged position, so the sweep records 58 inputs rather than one per frame; the test asserts at least the sweep's length.

Not done here

• Posting the strip and GIF on PHO-64: that publishes to the public repo, left to Miley. Both are in this workspace's .context/.
• The iOS simulator check (T12 · E, PHO-84), drawing with the light (T11), a Compose bump.

🤖 Generated with Claude Code