Quant tool: Consistent get_qdq_config and get_qnn_qdq_config behavior

[jambayk]

Description

Make the options and behavior for get_qdq_config and get_qnn_qdq_config consistent.

• Add calibration_providers and op_types_to_quantize options to get_qdq_config
• op_types_to_quantize takes precedence over nodes_to_exclude like in

  onnxruntime/onnxruntime/python/tools/quantization/base_quantizer.py

  Line 180 in 1be64f8

  if node.op_type not in self.op_types_to_quantize:

• Don't add a node's op types to the op type set if the node is excluded. This allows a caller to exclude an op type indirectly by excluding all node instances of the op type. This is useful since op_types_to_quantize is also used when creating the calibrator

  onnxruntime/onnxruntime/python/tools/quantization/quantize.py

  Line 703 in 1be64f8

  op_types_to_quantize,

  and there is no need to gather data for excluded ops.