fix(producer): drop empty trailing chunk slice in distributed render plan

[calcarazgre646]

The bug

resolveChunkPlan derives chunkCount from the naive count (min(maxParallelChunks, ceil(totalFrames / resolvedChunkSize))), then rounds effectiveChunkSize up to ceil(totalFrames / chunkCount). When that ceil rounds up, the first chunkCount - 1 chunks can already cover every frame, so buildChunkSlices emits a final slice with startFrame >= totalFrames — an empty [n, n) or inverted range.

renderChunk then rejects that slice (framesInChunk = endFrame - startFrame <= 0 → RenderChunkValidationError), and under Step Functions retries it exhausts retries and fails the whole distributed render, even though [0, totalFrames) was already fully covered. It also violates buildChunkSlices's own documented contract ("the union is exactly [0, totalFrames)").

Reachability

This is reachable straight from the user-facing CLI. --chunk-size and --max-parallel-chunks are first-class flags on hyperframes lambda render that flow into DistributedRenderConfig → resolveChunkPlan:

hyperframes lambda render --chunk-size 10 --max-parallel-chunks 12

on a ~4s / 30fps (121-frame) composition gives resolvedChunkSize=10, chunkCount=min(12, 13)=12, effectiveChunkSize=max(10, ceil(121/12)=11)=11. Slices 0–10 cover [0, 121); slice 11 is [121, 121).

The fix

Tighten chunkCount to ceil(totalFrames / effectiveChunkSize) after the size is finalized, so the trailing empty slice is dropped and the union stays exactly [0, totalFrames). This only ever lowers chunkCount in the explicit-small-chunkSize case; the auto-sized and large-chunkSize paths already satisfy ceil(totalFrames / effectiveChunkSize) >= chunkCount, so it's a no-op there (every existing test's asserted chunkCount is unchanged). maxParallelChunks is still respected.

Tests

plan.test.ts gains the 121 / 10 / 12 regression case (asserts chunkCount === 11 and the last slice is [110, 121)) plus a grid property test over a range of totalFrames × maxParallelChunks × explicit chunkSize asserting every slice is non-empty, contiguous from 0, and ends exactly at totalFrames.

[miguel-heygen]

Clean, correct fix. The root cause is well-explained: effectiveChunkSize rounding up post-cap left a [totalFrames, totalFrames) tail that renderChunk rejects, wedging the whole distributed render. The tighten-after-finalize approach is the right spot to fix it — no change to auto-sized or large-chunkSize paths.

The grid property test covering 150 combinations (6 frame counts × 5 max-parallel × 5 chunk sizes) is exactly the right test for an arithmetic invariant like this. No issues.