fix(paint): give the audio-reactive painters the scenario's clock

[LeadcodeDev]

Closes #201. Unblocks #202.

What

A waveform in a scene starting at t=73 s drew the analysis at 73 s into that scene. Same scene, same track, same build — only its position in the scenario differs:

scene position	trace height
first scene	42 px
starting at t≈73 s	0 px (flat)

PaintCtx carried no scenario-global time: time restarts at every scene, and frame_index is documented as "frames elapsed since this scene started". So waveform, audio_spectrum and the audio-reactive binding had nothing else to read, while AudioAnalysis is indexed on the scenario's own timeline. The two only agreed for a track at 0 and a scene at 0.

How

FrameTask carries global_frame — its index in the schedule, known at build time — and that becomes scenario_time on PaintFrame/PaintCtx and BuildAnimationCtx.

Deliberate boundaries:

• Everything else stays on time. A delay, a stagger, a reveal, a preset are all written against the scene. Only the three audio consumers move.
• A container's time_scale/time_offset does not remap it. That remaps the animation clock of a subtree; the audio plays at wall-clock speed regardless, so the scenario clock passes through untouched.
• The hit-map keeps the scene clock. It is a geometry probe for the studio overlay, never an encoded frame, and no painter reads the audio clock on that path.
• render_scene_bg_scaled did not get the parameter — the background layer paints no component.

The incremental encoder builds slots in isolation, so build_slot_frame_tasks takes the slot's start and renumbers rather than using the local Vec's length. slot_tasks_reproduce_the_full_builder_exactly caught that on the first run — exactly what it exists for.

Test

a_waveform_reads_the_scenario_clock_not_the_scene_clock: 3 s of sine then 1 s of silence, behind a 2 s filler scene. At frame 105 the scenario is at 3.5 s (silence) while the scene is 1.5 s old (still sine), so the two clocks disagree by construction.

Checked that it discriminates: reverting waveform.rs to ctx.time makes it fail, restoring the fix makes it pass. My first attempt at this test passed in both states — it was worthless, and I only found that out by trying to break it.

cargo test --workspace green, cargo fmt --check and cargo clippy --all-targets -- -D warnings clean.