ADR-0026: DPPolicy intra-device only + ShardSpec structural coords

DPPolicy no longer carries a cross-SIP axis. SIP-level placement is
solely controlled by torch.ahbm.set_device(rank) (ADR-0024); DPPolicy
itself describes only the cube × PE layout within one SIP. ShardSpec
switches to structural (sip, cube, pe) coordinates; the flat pe_index
field/property is fully removed — silent drift between global-flat and
SIP-local interpretations was a foot-gun flagged by ADR-0024 D11.

Breaking API (explicit TypeError / AttributeError):
- DPPolicy(sip=...) / DPPolicy(num_sips=...) -> TypeError
- ShardSpec.pe_index -> AttributeError
- ShardSpec(pe_index=...) -> TypeError
- resolve_dp_policy now takes target_sip= (required), no num_sips.

Downstream migration:
- PE allocator dict keyed by (sip, cube, pe) tuples, in both
  _ensure_allocators and _free_tensor. deploy_tensor uses tuple lookup.
- _create_tensor passes target_sip=current_sip; post-hoc pe_index
  shifting removed entirely.
- launch._compute_local_shape drops the dp.sip branch.
- Internal resolvers (column_wise / row_wise / replicate / tiled_*)
  return _LocalPeShard (cube-local identifier) instead of ShardSpec —
  resolve_dp_policy lifts them to full structural coords.

Tests:
- New tests/test_adr0026_dppolicy_intra_device.py (12 tests) pins the
  contract end-to-end.
- test_sip_parallel.py rewritten: SIP composition now modeled as two
  resolve_dp_policy(target_sip=...) calls (ADR-0024 launcher style).
- Call-site migration: test_tensor, test_va_integration, test_va_offset,
  test_runtime_api_tensor, test_tl_recv_async, test_ccl_* and benches
  gemm_single_pe, gpt3_qkv, va_offset_verify, ccl_allreduce (legacy
  branch) all use intra-device DPPolicy and structural ShardSpec.

Result: 523 passed, 1 strict xfail (ring_default_ws — unchanged
ADR-0024 Phase B blocker; architectural fix deferred to ADR-0027).

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

tests/test_va_integration.py

@@ -12,7 +12,7 @@ import pytest
 from kernbench.policy.address.allocator import AddressConfig, PEMemAllocator
 from kernbench.policy.address.pe_mmu import PeMMU
 from kernbench.policy.address.va_allocator import VirtualAllocator
-from kernbench.policy.placement.dp import column_wise, ShardSpec
+from kernbench.policy.placement.dp import DPPolicy, ShardSpec, resolve_dp_policy
 from kernbench.runtime_api.tensor import (
     TensorHandle,
     TensorShard,
@@ -37,9 +37,9 @@ _CFG = AddressConfig(
 )
 
 
-def _make_allocators(num_pe: int = 8) -> dict[int, PEMemAllocator]:
+def _make_allocators(num_pe: int = 8) -> dict[tuple[int, int, int], PEMemAllocator]:
     return {
-        i: PEMemAllocator(rack_id=0, sip_id=0, cube_id=0, pe_id=i, cfg=_CFG)
+        (0, 0, i): PEMemAllocator(rack_id=0, sip_id=0, cube_id=0, pe_id=i, cfg=_CFG)
         for i in range(num_pe)
     }
 
@@ -88,7 +88,11 @@ def test_deploy_tensor_assigns_va_base():
     """deploy_tensor with VA allocator assigns va_base to TensorHandle."""
     allocs = _make_allocators()
     va_alloc = _make_va_allocator()
-    placement = column_wise(shape=(1024, 512), itemsize=2, num_pe=8)
+    placement = resolve_dp_policy(
+        DPPolicy(cube="replicate", pe="column_wise"),
+        shape=(1024, 512), itemsize=2,
+        num_pe=8, num_cubes=1, target_sip=0,
+    )
 
     th = deploy_tensor(
         name="W",
@@ -107,7 +111,11 @@ def test_deploy_tensor_va_covers_all_shards():
     """VA allocation covers the entire tensor; each shard is at va_base + offset."""
     allocs = _make_allocators()
     va_alloc = _make_va_allocator()
-    placement = column_wise(shape=(1024, 512), itemsize=2, num_pe=8)
+    placement = resolve_dp_policy(
+        DPPolicy(cube="replicate", pe="column_wise"),
+        shape=(1024, 512), itemsize=2,
+        num_pe=8, num_cubes=1, target_sip=0,
+    )
 
     th = deploy_tensor(
         name="W",
@@ -128,7 +136,11 @@ def test_deploy_tensor_does_not_install_mmu_mappings():
     allocs = _make_allocators()
     va_alloc = _make_va_allocator()
     mmus = _make_mmus()
-    placement = column_wise(shape=(1024, 512), itemsize=2, num_pe=8)
+    placement = resolve_dp_policy(
+        DPPolicy(cube="replicate", pe="column_wise"),
+        shape=(1024, 512), itemsize=2,
+        num_pe=8, num_cubes=1, target_sip=0,
+    )
 
     deploy_tensor(
         name="W",
@@ -153,7 +165,7 @@ def test_tensor_va_property():
 
     allocs = _make_allocators(1)
     va_alloc = _make_va_allocator()
-    placement = [ShardSpec(pe_index=0, offset_bytes=0, nbytes=4096)]
+    placement = [ShardSpec(sip=0, cube=0, pe=0, offset_bytes=0, nbytes=4096)]
 
     t = Tensor(shape=(2048,), dtype="f16", name="test")
     t._handle = deploy_tensor(