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_install.py

@@ -0,0 +1,100 @@
+"""Tests for CCL backend install (ADR-0023 D10/D11)."""
+from __future__ import annotations
+
+from kernbench.ccl.install import (
+    install_ipcq,
+    linear_rank_to_pe,
+    load_ccl_config,
+    resolve_algorithm_config,
+)
+from kernbench.sim_engine.engine import GraphEngine
+from kernbench.topology.builder import resolve_topology
+
+
+def _engine():
+    topo = resolve_topology("topology.yaml").topology_obj
+    return GraphEngine(topo, enable_data=True), topo
+
+
+def test_load_ccl_config():
+    cfg = load_ccl_config()
+    assert "defaults" in cfg
+    assert "algorithms" in cfg
+
+
+def test_resolve_algorithm_config_default():
+    cfg = load_ccl_config()
+    merged = resolve_algorithm_config(cfg)
+    assert merged["algorithm"] == cfg["defaults"]["algorithm"]
+    # ccl.yaml no longer carries defaults.world_size — backend derives
+    # it from topology.yaml at install time. Just check the field is
+    # absent here (verified per-test where install_ipcq is called).
+    assert "world_size" not in merged or merged["world_size"] >= 1
+
+
+def test_resolve_algorithm_config_override():
+    cfg = load_ccl_config()
+    merged = resolve_algorithm_config(cfg, name="ring_allreduce_hbm")
+    assert merged["algorithm"] == "ring_allreduce_hbm"
+    assert merged["buffer_kind"] == "hbm"  # algo override
+    # defaults still apply
+    assert merged["n_slots"] == cfg["defaults"]["n_slots"]
+
+
+def test_linear_rank_to_pe():
+    engine, topo = _engine()
+    spec = topo.spec
+    # Cube 0 of SIP 0
+    assert linear_rank_to_pe(0, spec) == (0, 0, 0)
+    assert linear_rank_to_pe(7, spec) == (0, 0, 7)
+    # Should not exceed total PE count
+    pes_per_sip = (
+        spec["sip"]["cube_mesh"]["w"] * spec["sip"]["cube_mesh"]["h"]
+        * spec["cube"]["pe_layout"]["pe_per_corner"]
+        * len(spec["cube"]["pe_layout"]["corners"])
+    )
+    sips = spec["system"]["sips"]["count"]
+    total = sips * pes_per_sip
+    assert total >= 8
+
+
+def test_install_ipcq_neighbors_correct():
+    engine, topo = _engine()
+    cfg = load_ccl_config()
+    merged = resolve_algorithm_config(cfg, name="ring_allreduce_tcm")
+    # Force a single-cube 8-rank install for the assertions below.
+    merged["world_size"] = 8
+    plan = install_ipcq(engine, topo.spec, merged)
+
+    assert plan["world_size"] == 8
+    assert plan["buffer_kind"] == "tcm"
+
+    # Each rank should have E and W entries
+    for r, nbrs in plan["neighbor_table"].items():
+        assert "E" in nbrs
+        assert "W" in nbrs
+
+    # Inspect installed PE_IPCQ for rank 0
+    ipcq = engine._components["sip0.cube0.pe0.pe_ipcq"]
+    qp_e = ipcq.queue_pairs["E"]
+    qp_w = ipcq.queue_pairs["W"]
+    assert qp_e["peer"].pe == 1   # rank 0's E neighbor is rank 1
+    assert qp_w["peer"].pe == 7   # rank 0's W neighbor is rank 7
+    # rx_base addresses should be unique
+    assert qp_e["my_rx_base_pa"] != qp_w["my_rx_base_pa"]
+
+
+def test_install_ipcq_credit_stores_wired():
+    engine, topo = _engine()
+    cfg = load_ccl_config()
+    merged = resolve_algorithm_config(cfg, name="ring_allreduce_tcm")
+    merged["world_size"] = 8
+    install_ipcq(engine, topo.spec, merged)
+
+    # rank 0 (pe0) sending E goes to rank 1 (pe1)
+    # rank 0's peer_credit_store on E direction should equal rank 1's credit_inbox
+    pe0 = engine._components["sip0.cube0.pe0.pe_ipcq"]
+    pe1 = engine._components["sip0.cube0.pe1.pe_ipcq"]
+
+    qp_e = pe0.queue_pairs["E"]
+    assert qp_e["peer_credit_store"] is pe1.credit_inbox