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>

src/kernbench/runtime_api/tensor.py

@@ -72,7 +72,7 @@ def deploy_tensor(
     shape: tuple[int, ...],
     dtype: str,
     placement: list[ShardSpec],
-    allocators: dict[int, PEMemAllocator],
+    allocators: dict[tuple[int, int, int], PEMemAllocator],
     mem_kind: Literal["hbm", "tcm"] = "hbm",
     va_allocator=None,
 ) -> TensorHandle:
@@ -86,15 +86,15 @@ def deploy_tensor(
 
     shards: list[TensorShard] = []
     for spec in placement:
-        alloc = allocators[spec.pe_index]
+        alloc = allocators[(spec.sip, spec.cube, spec.pe)]
         if mem_kind == "hbm":
             pa = alloc.alloc_hbm(spec.nbytes)
         else:
             pa = alloc.alloc_tcm(spec.nbytes)
         shards.append(TensorShard(
-            sip=alloc._sip_id,
-            cube=alloc._cube_id,
-            pe=alloc._pe_id,
+            sip=spec.sip,
+            cube=spec.cube,
+            pe=spec.pe,
             pa=pa.encode(),
             nbytes=spec.nbytes,
             offset_bytes=spec.offset_bytes,
@@ -394,7 +394,8 @@ class Tensor:
     ) -> Tensor:
         """Set DP placement metadata (like torch.Tensor.to())."""
         if placement is None:
-            placement = [ShardSpec(pe_index=0, offset_bytes=0, nbytes=self.nbytes)]
+            placement = [ShardSpec(sip=0, cube=0, pe=0,
+                                   offset_bytes=0, nbytes=self.nbytes)]
         self._dp_metadata = DPMetadata(
             placement=placement, dp_policy=dp_policy,
             sip=sip, cube=cube, target_pe=target_pe,