Add PE-level IPCQ collective infra + unified ccl_allreduce bench (ADR-0023)

Major changes:

PE-level IPCQ infrastructure:
- New PE_IPCQ component: ring-buffer control plane with 4-direction
  neighbor mapping, head/tail pointers, backpressure (poll/sleep).
- PE_DMA extended with vc_comm channel for IPCQ outbound/inbound DMA,
  including in-flight data snapshot (D9) and op_log recording at
  outbound time for Phase 2 replay correctness.
- IpcqDmaToken piggyback model: data + metadata travel together,
  atomic visibility at receiver (invariant I6).
- Credit return fast path: bottleneck-BW latency, no fabric vc_comm.

Phase 2 data execution (ADR-0020 integration):
- op_log extended: DmaWriteCmd now captures src_space/src_addr for
  Phase 2 dma_write copy; ipcq_copy ops recorded at outbound time.
- DataExecutor replays dma_write + ipcq_copy in t_start order.
- Engine._flush_data_phase: incremental cursor-based replay after
  each engine.wait() so host reads see post-Phase-2 data.
- KernelRunner Phase 1 writes disabled when op_log is active to
  prevent stale data from corrupting the MemoryStore snapshot.

TLContext / kernel API:
- tl.send(dir, src=TensorHandle), tl.recv(dir, shape, dtype),
  tl.recv_async, tl.wait(RecvFuture), copy_to_dst mode.
- TensorHandle operator overloading (add/sub/mul/div) via thread-local
  active TLContext → MathCmd dispatch through PE_MATH.
- PE-local scratch allocator for math output handles.
- tl.load returns space="hbm" handles for correct Phase 2 addressing.
- Additional math functions: maximum, minimum, fma, clamp, softmax, cdiv.

Unified ccl_allreduce bench (PyTorch-compat host code):
- Single benches/ccl_allreduce.py with run() + worker(rank, ws, torch)
  split matching real PyTorch DDP worker pattern.
- torch.distributed facade: init_process_group, get_world_size,
  get_rank, get_backend, all_reduce, barrier — only real PyTorch names.
- AhbmCCLBackend: eager install_ipcq at init, all_reduce dispatches
  kernel via tensor shard metadata (n_elem from shards[0].nbytes).
- world_size derived from topology spec (sips × cubes × pes_per_cube)
  with optional algorithm-level override in ccl.yaml.

Tensor API (PyTorch-compat surface):
- Tensor.numpy(): gather-aware (all shards via VA-based addressing).
- Tensor.copy_(source): scatter from host tensor into sharded target.
- RuntimeContext.from_numpy(arr): host-side staging tensor.
- Tensor.data property fixed to use numpy() (was shards[0]-only).

Algorithm modules moved to src/kernbench/ccl/algorithms/:
- ring_allreduce, mesh_allreduce, tree_allreduce, hello_send.
- Each module exports kernel_args(world_size, n_elem) helper.
- ccl.yaml module paths updated to kernbench.ccl.algorithms.*.

Dead code removed:
- 7 per-variant bench files (ccl_allreduce_{tcm,hbm,sram}, etc.).
- _run_ccl_bench greenlet-per-SIP scheduler.
- benches.loader.is_ccl_bench + run_rank detection.
- benches/ccl/ directory.

Tests:
- New test_ccl_allreduce_matrix.py: 7 parametrized cases
  (ring×3 buffers, ring 8/16, mesh 4, tree 7).
- New test_runtime_api_tensor.py: copy_/numpy/from_numpy unit tests.
- Existing tests updated for new import paths + world_size_override.

Docs:
- Korean ccl-author-guide.md and ADR-0023 paths updated.
- New English versions: ccl-author-guide.en.md, ADR-0023.en.md.

502 tests pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/test_ccl_deadlock_detection.py

@@ -0,0 +1,125 @@
+"""Tests for IPCQ deadlock detection (ADR-0023 D14 F3)."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+import pytest
+import simpy
+
+from kernbench.ccl import diagnostics
+from kernbench.common.ipcq_types import (
+    IpcqEndpoint,
+    IpcqInitEntry,
+    IpcqRecvCmd,
+    IpcqRequest,
+)
+from kernbench.components.builtin.pe_ipcq import PeIpcqComponent
+from kernbench.runtime_api.kernel import IpcqInitMsg
+from kernbench.topology.types import Node
+
+
+@dataclass
+class _FakeTxn:
+    request: Any
+    done: simpy.Event
+    result_data: dict[str, Any] = field(default_factory=dict)
+
+
+def _make_isolated_pe_ipcq(env):
+    node = Node(
+        id="sip0.cube0.pe0.pe_ipcq", kind="pe_ipcq",
+        impl="builtin.pe_ipcq", attrs={}, pos_mm=None,
+    )
+    comp = PeIpcqComponent(node, ctx=None)
+    comp.in_ports["host"] = simpy.Store(env)
+    comp.out_ports["sip0.cube0.pe0.pe_dma"] = simpy.Store(env)
+    comp.start(env)
+
+    peer_credit = simpy.Store(env)
+    ep = IpcqEndpoint(
+        sip=0, cube=0, pe=1, buffer_kind="tcm",
+        rx_base_pa=0x10_000, rx_base_va=0,
+        n_slots=4, slot_size=4096,
+    )
+    init_msg = IpcqInitMsg(
+        correlation_id="t", request_id="t",
+        target_sips=(0,), target_cubes=(0,), target_pe=0,
+        entries=(IpcqInitEntry(
+            direction="W", peer=ep,
+            my_rx_base_pa=0x40_000, my_rx_base_va=0,
+            n_slots=4, slot_size=4096,
+            peer_credit_store=peer_credit,
+        ),),
+        backpressure_mode="sleep",
+        buffer_kind="tcm",
+        credit_size_bytes=16,
+    )
+    done = env.event()
+    comp.in_ports["host"].put(_FakeTxn(request=init_msg, done=done))
+    env.run(until=done)
+    return comp
+
+
+def test_pointer_dump_includes_blocked_state():
+    """A blocked recv should still be visible in the pointer dump."""
+    env = simpy.Environment()
+    comp = _make_isolated_pe_ipcq(env)
+
+    # Issue a recv that will block (no data has arrived)
+    recv_cmd = IpcqRecvCmd(direction="W", shape=(8,), dtype="f16", handle_id="r1")
+    req = IpcqRequest(command=recv_cmd, done=env.event())
+    comp.in_ports["host"].put(req)
+    env.run(until=10)
+    assert not req.done.triggered
+
+    # Pointer dump should show my_tail=0 and peer_head_cache=0
+    # We need to use the engine API but for an isolated component, just call directly
+    class FakeEngine:
+        _components = {"sip0.cube0.pe0.pe_ipcq": comp}
+
+    dump = diagnostics.pointer_dump(FakeEngine())
+    assert "my_tail=0" in dump
+    assert "peer_head_cache=0" in dump
+
+
+def test_deadlock_detection_recv_without_send():
+    """A recv with no matching sender → SimPy schedule empties → engine
+    raises ``IpcqDeadlock`` with a pointer dump.
+    """
+    from kernbench.ccl.diagnostics import IpcqDeadlock
+    from kernbench.policy.placement.dp import DPPolicy
+    from kernbench.runtime_api.bench_runner import run_bench
+    from kernbench.runtime_api.types import resolve_device
+    from kernbench.sim_engine.engine import GraphEngine
+    from kernbench.topology.builder import resolve_topology
+
+    def deadlock_kernel(t_ptr, n_elem, tl):
+        # Every PE just receives, no sends → no one delivers → deadlock
+        tl.recv(dir="W", shape=(n_elem,), dtype="f16")
+
+    topo = resolve_topology("topology.yaml")
+
+    def run(torch):
+        torch.install_ipcq(
+            algorithm="ring_allreduce_tcm", world_size_override=8,
+        )
+        a = torch.zeros(
+            (1, 8 * 8),
+            dtype="f16",
+            dp=DPPolicy(
+                sip="replicate", cube="replicate", pe="column_wise",
+                num_sips=1, num_cubes=1,
+            ),
+            name="dl_in",
+        )
+        torch.launch("dl", deadlock_kernel, a, 8)
+
+    with pytest.raises(IpcqDeadlock):
+        run_bench(
+            topology=topo, bench_fn=run,
+            device=resolve_device("all"),
+            engine_factory=lambda t, d: GraphEngine(
+                getattr(t, "topology_obj", t), enable_data=True
+            ),
+        )