Add virtual memory support: PE_MMU, VA allocator, fabric MmuMapMsg

Implement VA/MMU layer (ADR-0011 Phase 1) enabling Triton kernels to use
contiguous virtual addresses on sharded tensors.

Key changes:
- PE_MMU component: hybrid inbox (MmuMapMsg) + sync translate() for PE_DMA
- VirtualAllocator + PEMemAllocator: free-list with coalescing
- MmuMapMsg/MmuUnmapMsg fabric path with SIP-level routing
- DPPolicy-based mapping: replicate=local, sharded=broadcast
- Tensor lifecycle: del + weakref cleanup, context manager
- Rename: TensorHandle.pa→addr, DmaReadCmd.src_pa→src_addr, ctx→torch

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

src/kernbench/common/pe_commands.py

@@ -28,7 +28,7 @@ class TensorHandle:
     """
 
     id: str
-    pa: int                          # physical address in HBM/TCM
+    addr: int                        # address (VA when MMU enabled, PA otherwise)
     shape: tuple[int, ...]
     dtype: str
     nbytes: int                      # total byte size
@@ -50,19 +50,19 @@ class CompletionHandle:
 
 @dataclass(frozen=True)
 class DmaReadCmd:
-    """DMA READ: HBM → PE_TCM."""
+    """DMA READ: HBM → PE_TCM. src_addr is VA (translated to PA by PE_DMA)."""
 
     handle: TensorHandle
-    src_pa: int
+    src_addr: int
     nbytes: int
 
 
 @dataclass(frozen=True)
 class DmaWriteCmd:
-    """DMA WRITE: PE_TCM → HBM."""
+    """DMA WRITE: PE_TCM → HBM. dst_addr is VA (translated to PA by PE_DMA)."""
 
     handle: TensorHandle
-    dst_pa: int
+    dst_addr: int
     nbytes: int
 
 
@@ -108,7 +108,7 @@ class CompositeCmd:
     op: Literal["gemm", "math"]
     a: TensorHandle
     b: TensorHandle | None
-    out_pa: int
+    out_addr: int
     out_nbytes: int
     math_op: str | None = None       # for op="math": which math operation