Merge changes from topics "syncexec-ndk-shallow-aosp", "replace_asymm", "fp16-op-add"
* changes:
Implement ANeuralNetworksExecution_compute() and add related properties.
Adds float16 support for CONCATENATION.
Add float16 support for SPLIT
Fix TRANSPOSE op.
Adds float16 support for LOGISTIC/SOFTMAX.
Adds float16 support for RELU*.
Adds float16 support for MUL/DIV.
Add float16 support for TOPK_V2 op
Replace TENSOR_QUANT16_ASYMM with TENSOR_QUANT16_SYMM
Refactor HAL version validation
Adds float16 support for ADD operation.
Implement implicit float16 conversion in test generator.
Add Float16 support to SUB
Add layer normalization support to LSTM op
Add Float16 support to TANH
Add Float16 support to MAXIMUM and MINIMUM
Add Float16 support to GATHER
Add Float16 support to CAST
Add Float16 support to TILE
Fix generated test names for Float16 variations
Fix TENSOR_FLOAT16 support in CTS generator
Replace NN_CHECK with more flexible NN_RET_CHECK
Add synchronous execution API ANeuralNetworksExecution_compute().
Adds float16 support to generated tests.
Autogenerates VTS ValidationTest tests.
