* ROIAlign op skeleton

* Expose ROIAlign via opset2

* ROIAlign attribute visitor

* Batch indices input for ROIAlign

* Shape inference for ROIAlign

* ROIAlign mode attribute validation

* onnx_importer ROIAlign op

* size_t of get_attribute_value in onnx_importer::Node

* Explicit cast size_t -> int64_t when creating Dimensions

* ...aaand some new lines to make clang format happy

* Rank check fix for the batch indices input.
Co-Authored-By: Adam Osewski <adam.osewski@intel.com>

* Typo
Co-Authored-By: Adam Osewski <adam.osewski@intel.com>

* Empty opset3 definition

* Move ROIAlign to opse3

* Review comments

* Correct rank check for batch indices input

* Basic shape inference test for ROIAlign

* Move ROIAlign to ops from experimental/layers

* ROIAlign -> RoiAlign
Co-Authored-By: Katarzyna Mitrus <katarzyna.mitrus@intel.com>

* Move ROIAlign from v0 to v3

* Support more data types in ROIAlign

* Move the ROIAlign from opset2_tbl to opset3_tbl

* Don't include opset2 in opset3.hpp

* PoolingMode enum for ROIAlign attribute

* Use EnumNames to handle string/enum conversion

* More checks to prevent segfaults

* Remove the std prefix for string
Co-Authored-By: Adam Osewski <adam.osewski@intel.com>

* Fix more potential segfaults

* Build break fix

* ROIAlign serializer

* PR feedback

* Disable the ROIAlign op in GPU backend

* ROIAlign type prop UT

* More shape inference UT

* PR comments for RoiAlign

* code formatting
Co-authored-by: Adam Osewski <adam.osewski@intel.com>
Co-authored-by: Katarzyna Mitrus <katarzyna.mitrus@intel.com>
Co-authored-by: Robert Kimball <robert.kimball@intel.com>