…graph runtime

Implement register-based communication between AICPU and AICore for
a2a3 platform with host_build_graph runtime, using hardware SPR access
for task dispatch and status polling.

Architecture:
- Platform layer (a2a3): Register access primitives and address retrieval
- Runtime layer (host_build_graph): Task dispatch and status polling logic

Platform changes (a2a3):
- Add AICore register read/write APIs (aicpu_regs.h/cpp, host_regs.h/cpp)
- Implement register address retrieval via CANN HAL (get_aicore_regs)
- Add register configuration constants in platform_config.h
- Support physical_core_id tracking for register mapping

Runtime changes (host_build_graph):
- Replace memory-based handshake with register-based dispatch
- Use DATA_MAIN_BASE register for task_id dispatch (encoding: task_id+1)
- Use COND register for status polling (IDLE=0, BUSY=1)
- Add Runtime.regs field for register address array
- Track executing_task_ids[] in AICPU executor