HLS Layernorm

[auphelia]

This PR integrates the HLS implementation of LayerNorm (from finn-hlslib). The HLS op accepts integer/quantized inputs and float inputs.

Selection behavior

• Quantized inputs: the compiler automatically derives and selects the HLS LayerNorm.
• Float inputs: selection depends on the target device.
  • Versal: RTL LayerNorm remains the default (uses DSP58).
  • Non‑Versal: HLS LayerNorm is selected (devices without DSP58).
• You can explicitly prefer HLS on any target by setting preferred_impl_style="hls" on the HWCustomOp.

Notes

• Output type remains float for both implementations.