Function draw_sprites_parallel 

pub fn draw_sprites_parallel(
    target: DrawTarget<'_>,
    cam: &CameraState,
    settings: &OpticastSettings,
    lighting: &SpriteLighting<'_>,
    sprites: &[Sprite],
) -> u32

Draw a sprite into a framebuffer + z-buffer.

Top-level dispatcher mirroring voxlap5.c:9818-9828:

• Skips on flags & INVISIBLE.
• Skips on flags & KFA (animation path; out of scope for R6).
• Skips on flags & NO_Z (handled by drawboundcubenozsse, not yet ported — the four oracle sprite poses all use z-tested rendering).

Otherwise: cull → setup math → 9-arm per-voxel iteration → per-voxel rasterize via the R6.4 drawboundcubesse port.

Returns the total number of pixels written across all voxels of the sprite (== sum of z-test passes). Zero means the sprite produced no visible pixels (culled, fully behind near plane, or totally occluded). Render a batch of sprites in parallel via rayon::par_iter.

Each sprite runs its own draw_sprite pass on its own thread, writing to the shared DrawTarget (raw pointers; `Copy + Send

• Sync`) under the z-test arbitration contract: a pixel write only fires when the new sprite’s z is strictly less than the current zbuffer value. For non-overlapping sprites the writes are pairwise-disjoint and the output is byte-identical to a sequential pass over the same sprite list. For overlapping pixels, two sprites at exactly tied z-values produce a non-deterministic last-writer-wins outcome — visually indistinguishable but hash-non-deterministic.

Returns the sum of draw_sprite return values (total pixels written across all sprites).

RAYON_NUM_THREADS=1 (or no parallelism worth) ⇒ effectively sequential; rayon falls back to running each closure on the calling thread without contention.

Use this for engine scenes with dozens-to-hundreds of sprites; the per-sprite overhead amortises well past ~4 sprites on consumer-class hardware.