[BUG] MemPoolSplitMHSAPass produces shapeless tensors

[lukamac]

Description

After merging the #128, shape inference will be disallowed and with it, passes that produce shapeless tensors.
In the current Deeploy codebase, I have stumbled upon only one such pass, and that is the MemPoolSplitMHSAPass which gets triggered on the Tests/ICCT_ITA_8 test.
The aforementioned PR disables that test until the Pass gets fixed and this issue follows that task.

To Reproduce

Run the following command from DeeployTest:

python testRunner_mempool.py -t Tests/ICCT_ITA_8

and Deeploy will shout at you.

To confirm that it is MemPoolSplitMHSAPass that creates shapeless tensors, I have created the following TopologyOptimizer:

class MemPoolTopologyOptimizer(TopologyOptimizer):

    def optimize(self, graph: gs.Graph) -> Tuple[gs.Graph]:
        shapeless = set(t.name for t in graph.tensors().values() if t.shape is None)
        if len(shapeless) > 0:
            print(f"Graph already has some shapeless tensors: {shapeless}")
        for i, _pass in enumerate(self.passes):
            prevShapeless = shapeless
            graph = _pass.apply(graph)
            graph.cleanup().toposort()
            shapeless = set(t.name for t in graph.tensors().values() if t.shape is None)
            newShapeless = shapeless - prevShapeless
            if len(newShapeless) > 0:
                print(f"The {i}th pass {type(_pass).__name__} produced tensors without shape. Tensors: {newShapeless}")
        breakpoint()
        return graph

and used it instead of the normal optimizer for the MemPoolOptimizer in Platform.py.

Expected behavior

Deeploy should not shout at you anymore.