Merge changes from topics "quant8_symm", "conv2d"

* changes:
  Replace ANNOperandType::ExtraParams with a function call.
  Fix tests, add depth_multiplier value check in DEPTHWISE_CONV_2D.
  Add new op INSTANCE_NORMALIZATION.
  Modify ROI_ALIGN specification.
  Modify ROI_POOLING specification.
  DEPTHWISE_CONV_2D with per-channel quantization.
  Typedef bool8 in OperationUtils
  Fix quant8 input validation bug in CONCATENATION
  Add EMBEDDING_LOOKUP_SPARSE op
  Add missing ANeuralNetworksOperandType.extraParams in ExecutionStep::addOperand
  Remove unneeded header
  Add support for QUANT8_SYMM_PER_CHANNEL graph inputs
  CONV_2D with per-channel quantization.
  Adds float16 support for SVDF.
  Adds float16 support for RNN.
  Fixes seed input type.
  Add SPARSE_TO_DENSE op
  Add Float16Scalar to test generator
  Fix OperandExtraParamsTest failures
  Add new ops: SIN, LOG, SQRT, RSQRT
  Add TENSOR_QUANT8_SYMM_PER_CHANNEL to operand types.
  Add new op EXP
  Add new op NEG