Add logical objFifo placeholder op for connection reuse

[jtuyls]

Adds the amdaie.logicalobjectfifo.placeholder operation that represents a logical objectFifo to be filled in later. This enables reuse of connections/circular DMAs/physical AIE channels for different data packets, which helps with (fused) operations with more than 2 inputs. This is especially useful for reading from/writing to DDR.

Before this PR, the circular DMA connections would be always be tied to a HAL buffer on DDR/L3 and the control code NPU instruction could only change the addressing and not the input buffer:

%0 = hal.interface.binding.subspan layout(#pipeline_layout) set(0) 
  binding(0) alignment(64) offset(%c0) flags(ReadOnly) : memref<1024xi32>
%obj0 = amdaie.logicalobjectfifo.from_memref %0, {%tile_0_0} 
  : memref<1024xi32> -> !amdaie.logicalobjectfifo<memref<1024xi32>>
%1 = memref.alloc() : memref<1024xi32, 1 : i32> 
%obj1 = amdaie.logicalobjectfifo.from_memref %1, {%tile_0_1} 
  : memref<1024xi32, 1 : i32> -> !amdaie.logicalobjectfifo<memref<1024xi32, 1 : i32>>
%connection = amdaie.circular_dma_cpy_nd(%obj1[] [] [], %obj0[] [] []) 
  : (!amdaie.logicalobjectfifo<memref<1024xi32, 1 : i32>>, !amdaie.logicalobjectfifo<memref<1024xi32>>)
amdaie.controlcode {
      %npu_dma = amdaie.npu.dma_cpy_nd %connection([] [] [], [%c0, %c32] [%c32, %c32] [%c32, %c1])
      amdaie.end
    }

By introducing placeholders, the NPU dma operation can specify the actual logical objFifo to be used:

%0 = hal.interface.binding.subspan layout(#pipeline_layout) set(0) 
  binding(0) alignment(64) offset(%c0) flags(ReadOnly) : memref<1024xi32>
%1 = memref.alloc() : <1024xi32, 1 : i32>
%obj1 = amdaie.logicalobjectfifo.from_memref %1, {%tile_0_1}
  : memref<1024xi32, 1> -> !amdaie.logicalobjectfifo<memref<1024xi32, 1>>
%ph = amdaie.logicalobjectfifo.placeholder{} 
  : !amdaie.logicalobjectfifo<memref<2048xi32>>
%connection = amdaie.circular_dma_cpy_nd(%obj1[] [] [], %ph[] [] [])
  : (!amdaie.logicalobjectfifo<memref<1024xi32, 1>>, !amdaie.logicalobjectfifo<memref<1024xi32>>)
amdaie.controlcode {
  %obj0 = amdaie.logicalobjectfifo.from_memref %0, {%tile_0_0}
     : memref<1024xi32> -> !amdaie.logicalobjectfifo<memref<1024xi32>>
   %npu_dma = amdaie.npu.dma_cpy_nd %connection([] [] [],  %obj0[%c0, %c32] [%c32, %c32] [%c32, %c1]) 
     : source_type = !amdaie.logicalobjectfifo<memref<1024xi32>>
   amdaie.end
}

[Abhishek-Varma]

Great! A few small comments from my end. Not blocking this.