PR #62817: [onednn] Enable auto_mixed_precision for fp16 on cpu

Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/62817

This PR enables AMP FP16 on supporting Intel Xeon CPUs. If there is no GPU available to run FP16, it checks if FP16 can be executed on CPU.
Use config `auto_mixed_precision`  to enable FP16 on CPUs.
Copybara import of the project:

--
fc1ee3bc by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

[onednn] Enable auto_mixed_precision for fp16 on cpu

--
9a3c8d59 by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

Address review comments - update comments as per guidelines.

--
27b1a962 by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

Address review comments

--
9a3ef902 by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

Address review comments

--
0ab5b80f by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

minor change

--
182a1738 by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

Address review comments

--
9780d75a by Gauri1 Deshpande <gauri1.deshpande@intel.com>:

Address re-review comments

Merging this change closes #62817

FUTURE_COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/tensorflow/pull/62817 from Intel-tensorflow:gaurides/amp_fp16 9780d75a
PiperOrigin-RevId: 615915563