Add SchedulerV2 (pe_accel), DPPolicy overrides, and new benchmarks

- Add cycle-accurate PE accelerator scheduler (SchedulerV2) with tiled
  GEMM/Math pipelines (DMA_IN → GEMM → MATH → DMA_WB)
- Add DPPolicy num_pes/num_cubes/num_sips overrides for single-PE testing
- Support tuple target_pe for targeting specific PE subsets
- Add gemm_single_pe and gpt3_qkv benchmarks
- Switch default topology to pe_scheduler_v2

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

src/kernbench/components/builtin/m_cpu.py

@@ -320,10 +320,12 @@ class MCpuComponent(ComponentBase):
         else:
             txn.done.succeed()
 
-    def _resolve_pe_ids(self, target_pe: int | str) -> list[int]:
+    def _resolve_pe_ids(self, target_pe: int | tuple | str) -> list[int]:
         """Return list of PE IDs to fan out to (used by kernel launch fan-out)."""
         if isinstance(target_pe, int):
             return [target_pe]
+        if isinstance(target_pe, tuple):
+            return list(target_pe)
         # "all": all PEs in local cube
         n_slices = 8
         if self.ctx and self.ctx.spec: