fix(producer): reject impossible distributed durations

[miguel-heygen]

What

Distributed planning now fails fast when a probed composition duration would produce an impossible frame count. Instead of freezing a plan and fanning out chunk work for an unbounded timeline sentinel, plan() raises a typed DISTRIBUTED_DURATION_OUT_OF_RANGE validation error before chunk metadata is written.

Why

A composition can accidentally expose an unbounded runtime timeline as a huge finite duration, for example from an infinite animation loop without an explicit finite root duration. In distributed mode that turns into billions of frames and pushes the failure downstream into chunk workers and orchestration. The producer should reject that deterministic authoring/configuration error at the planning boundary.

How

Added a shared distributed-duration validator with a one-day operational ceiling and wired it immediately after browser probing, before plan freeze and chunk fanout. The typed error code is exported from @hyperframes/producer/distributed so adapters can classify it as terminal while continuing to retry real infrastructure faults.

Test plan

• Unit tests added/updated
• Manual testing performed
• Documentation updated (if applicable)

Verification run:

• bun test packages/producer/src/services/render/planValidation.test.ts packages/producer/src/services/distributed/planSizeCap.test.ts packages/producer/src/services/distributed/publicExports.test.ts --timeout 60000
• bun run format:check
• bun run --filter @hyperframes/producer typecheck
• PATH="/tmp/node22-bin:$PATH" bun run lint
• bun run build:producer

---