- SameResize_Batch hlslib op always pads with 0, i/odt must support
zero values
- Added assertion to check for PaddingStyle=2, the only tested
value

This reverts commit 2e96d065.

* use make_new_valueinfo_name instead of own routine
* set dtype on newly created tensor

[Transform] Add checks to AbsorbAddIntoMultiThreshold and AbsorbMulIntoMultiThreshold transforms to work only on linear segments of graphs with branches

[TRANSFORM] Generalize from past scalar to past any linear op (this was supported, but restricted). Fix node sorting problem, by iterating over a copy list of the node instead of iteration over graph.node. Fix corner case when both branches are empty and come from the same node

LabelSelect is currently broken for signed inputs
will open issue separately

[Transform] Add checks to collapse repeated transform to work only on linear segments of graphs with branches

[HLSCustomOp] Add first draft of HLSCustomOp that corresponds to finn-hlslib SameResize_Batch function

[Transform] Add checks to reorder transforms to work only on linear segments of graphs with branches