Pt2e workflow for torch >= 2.5

[shivadbhavsar]

The current set of calls:

torch.export.export -> prepare_pt2e -> convert_pt2e -> torch.compile

has different behavior then previously. The quantization params are lifted to inputs (rather than being static parameters) by torch.compile and there is no simple switch for disabling this behavior.

Requires further investigation and potential rework for the pt2e workflow in general (possibly to just directly convert exported model wihtout using torch.compile). This needs a few additional converters and updates in lowering passes.