Nightly tests failing around sparse COO

[woodsp-ibm]

Sparse, before version 0.16.0 eg in 0,15.5 the COO constructor allowed None for shape and in torch connector that default is being used

qiskit-machine-learning/qiskit_machine_learning/connectors/torch_connector.py

Line 240 in de2cf2c

grad_coo = COO(grad_output.indices(), grad_output.values())

This resulted then in this in the logs

/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/sparse/_coo/core.py:245: DeprecationWarning: shape should be provided. This will raise a ValueError in the future.

Version 0.16.0 has now been released and it now raises a ValueError causing tests to fail. shape is now required.

      File "D:\a\qiskit-machine-learning\qiskit-machine-learning\qiskit_machine_learning\connectors\torch_connector.py", line 240, in backward
    grad_coo = COO(grad_output.indices(), grad_output.values())
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

      File "C:\hostedtoolcache\windows\Python\3.12.10\x64\Lib\site-packages\sparse\numba_backend\_coo\core.py", line 245, in __init__
    raise ValueError("`shape` was not provided.")

    ValueError: `shape` was not provided.

While we could pin sparse to < 0.16 I think the code should be changed to pass shape in the COO construction as is now required,