Refactor: drop Dist prefix from runtime types, files, and constants

[ChaoWao]

Summary

Mechanical rename sweep across the hierarchical runtime — no behavioural changes. Final PR in the refactor chain (follows #563, #564, #570, #572, #575, #578, #583).

File renames (history preserved via git mv)

• src/common/distributed/dist_*.{h,cpp} → *.{h,cpp}
• tests/ut/cpp/test_dist_*.cpp → test_*.cpp
• tests/ut/py/test_dist_worker/ → test_worker/
• python/bindings/dist_worker_bind.h → worker_bind.h

Type renames

DistWorker → Worker, DistOrchestrator → Orchestrator, DistScheduler → Scheduler, DistWorkerManager → WorkerManager, DistRing → Ring, DistScope → Scope, DistTensorMap → TensorMap, DistReadyQueue → ReadyQueue, DistSubmitResult → SubmitResult, DistTaskSlotState → TaskSlotState, DistAllocResult → AllocResult.

Constant renames

DIST_MAX_RING_DEPTH → MAX_RING_DEPTH, DIST_MAILBOX_SIZE → MAILBOX_SIZE, DIST_HEAP_ALIGN → HEAP_ALIGN, DIST_INVALID_SLOT → INVALID_SLOT, DIST_MAX_SCOPE_DEPTH → MAX_SCOPE_DEPTH, DIST_ALLOC_TIMEOUT_MS → ALLOC_TIMEOUT_MS, DIST_DEFAULT_HEAP_RING_SIZE → DEFAULT_HEAP_RING_SIZE, DIST_MAILBOX_ARGS_CAPACITY → MAILBOX_ARGS_CAPACITY.

Python binding naming

The C++ types Worker and Orchestrator are bound as _Worker and _Orchestrator to avoid colliding with the user-facing Python simpler.worker.Worker factory and simpler.orchestrator.Orchestrator facade, which wrap them. Matches the existing _ChipWorker → ChipWorker pattern.

Roadmap cleanup

Removes three stale roadmap docs now that the refactor chain is complete:

• docs/roadmap.md — every bullet is now covered by the per-component docs (orchestrator/scheduler/worker-manager/task-flow). The "Behavioural notes" content (release_ref +1 threshold, fork hygiene setenv + pthread_atfork) already lives in orchestrator.md §8 and §"Fork hygiene before fork". Also removes six forward-reference blockquotes ("see roadmap.md for landed-vs-planned") from the per-component docs and the README doc table entry.
• src/{a2a3,a5}/runtime/tensormap_and_ringbuffer/docs/ROADMAP.md — all three proposed features have landed, via a different API than the roadmap described. MixedKernels + atomic cluster dispatch replaced the allocate_cluster/free_cluster proposal (explicitly "Out of Scope" in SUBMIT_BY_CLUSTER.md); PTO2LaunchSpec covers SPMD→MPMD block_incore expansion; cube+vector co-scheduling rides on the MixedKernels path. Current runtime design lives in RUNTIME_LOGIC.md + SUBMIT_BY_CLUSTER.md.

Test plan

• pip install -e . builds cleanly
• Python unit tests: 160 passed, 5 skipped (3 pre-existing failures require torch, unrelated)
• C++ unit tests: 7/7 passed via ctest (test_tensormap, test_ring, test_scope, test_orchestrator, test_scheduler, test_a2a3_pto2_fatal, test_a5_pto2_fatal)
• grep -rnE "Dist[A-Z]|DIST_" src/ python/ returns no results
• ls src/common/distributed/dist_* / ls tests/ut/cpp/test_dist_* return nothing
• Hardware tests (no device available in authoring environment)

[gemini-code-assist]

Code Review

This pull request implements a major refactor of the distributed runtime by dropping the "Dist" prefix from class names, constants, and source files, and removing the project roadmap and associated status documentation. Feedback highlights a significant risk of name collisions resulting from moving generic names into the global namespace, suggesting the use of a proper C++ namespace for these definitions. Furthermore, inconsistencies were noted in the documentation updates, specifically regarding the naming of configuration types.