Refactor: drop WorkerPayload; IWorker::run takes TaskArgsView

[ChaoWao]

Summary

• IWorker::run(uint64_t callable, TaskArgsView args, ChipCallConfig) replaces the WorkerPayload dispatch carrier. Each worker interprets callable per its semantics (ChipCallable ptr / registry id / orch-fn handle).
• DistTaskSlotState drops chip_storage_list in favour of TaskArgs (single) / task_args_list (N-SPMD group) + is_group_ flag. slot.args_view(i) is a zero-copy TaskArgsView. callable_ptr renames to callable.
• view_to_chip_storage moves from submit into ChipWorker::run — the L2 ABI POD is assembled lazily at the worker.
• PROCESS-mode mailbox writes a length-prefixed blob via write_blob; the Python child decodes it through a new _ChipWorker.run_from_blob binding that rebuilds the ChipStorageTaskArgs POD on the child heap.
• WorkerThread queue now carries WorkerDispatch { slot, group_index }; the thread resolves slot state via ring->slot_state(id). Scheduler pushes N dispatches per group member.
• Strict-1 per-scope rings were split out into a separate PR-Scope (plan revision 2026-04-15) that also exposes user-facing scope_begin / scope_end. PR-C intentionally does not touch ring partitioning.

Docs

docs/roadmap.md, docs/scheduler.md, docs/worker-manager.md, docs/task-flow.md, and the PR-C entry in the plan are updated to match the new dispatch surface, slot shape, and PR-Scope split.

Test plan

• Python unit tests pass: pytest tests/ut/py/test_dist_worker/ (18/18)
• C++ unit tests pass: ctest in build/ut_cpp (7/7 — ring, scope, tensormap, orchestrator, scheduler, pto2_fatal × 2)
• pip install --no-build-isolation -e . builds cleanly
• grep -r WorkerPayload src/ python/ returns only historical doc references (all source clean)
• grep -r chip_storage_list src/ python/ returns 0 matches
• Hardware / sim scene tests (not run locally — need device)

[gemini-code-assist]

Code Review

This pull request refactors the task dispatch system by removing the WorkerPayload carrier and updating the IWorker::run interface to use TaskArgsView. Task arguments are now stored directly in the DistTaskSlotState, with the conversion to the L2 ABI ChipStorageTaskArgs POD deferred to the worker's execution phase. The WorkerThread now utilizes a WorkerDispatch handle to resolve task details from the DistRing at dispatch time. Documentation and tests have been updated to reflect these architectural changes. Feedback was provided regarding the efficiency of buffer allocation within the Python worker loop, suggesting that moving the allocation outside the loop or avoiding the copy could reduce garbage collector pressure.