webnn: Support int4 dequantization and per-channel quantization
1, Emulate the quantize operation whose calculation follows the
expression clamp(tfl.round(input / scale) + zeroPoint, 0, 255).
2, Support pre-channel quantization and dequantization
3, Support int4 dequantization, but quantization operation doesn't
support int4 for tfl.quantize operator and it can't be emulated
because the cast operation also doesn't support int4.
Bug: 375614290, 376722724
Change-Id: Iccbf69d8426e5729b35455357c99f00effaa21e2
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6015965
Commit-Queue: Junwei Fu junwei.fu@intel.com
Reviewed-by: ningxin hu ningxin.hu@intel.com
Reviewed-by: Reilly Grant reillyg@chromium.org
Cr-Commit-Position: refs/heads/main@{#1385963}