Support int8 / qs8 `AVERAGE_POOL_2D`

Average pooling is used in many neural networks so it would be great if XNNPACK would support a `int8` / `qs8` version of this operation as well.

Currently `int8` max pooling is already supported, but only a [`uint8` version of average pooling](https://github.com/google/XNNPACK/blob/db977bf482436fadddfcb1096329761cc634c8ee/src/operators/average-pooling-nhwc.c#L45) seems to be implemented. It would be great be able to run int8 average pooling via XNNPACK as well given that it is the new default quantization in TFLite.