API: different shape between `matrix.sum()` (return scaler) and `matrix@matrix` (return matrix)

[Zeroto521]

The result of one_dim_matrix.sum() is Expr (scaler). It was changed to a scaler if the size of the result is 1.
But the result of matrix@matrix is MatrixVariable (matrix). The size of the result matrix@matrix is 1.
Same shape but different result. Should we keep the same?

• Use numpy style, don't change the result at all. Keep the result as the matrix.
• If the size is one, change it to scaler.

In [1]: from pyscipopt import Model

In [2]: model = Model()

In [3]: x = model.addMatrixVar(3)

In [4]: y = model.addMatrixVar(3)

In [5]: type(x.sum())  # the size is 1, and result is a scaler
Out[5]: pyscipopt.scip.Expr

In [6]: type(x @ y)  # the size is 1, but result is a matrix
Out[6]: pyscipopt.scip.MatrixVariable

In [7]: (x @ y).shape
Out[7]: ()

In [8]: type((x @ y).item())
Out[8]: pyscipopt.scip.Expr

[Zeroto521]

The matrix was added in version 5.4.1. The list is still used as the primary data source. The matrix-related behaviors differ from those of NumPy in some ways, such as #1034.

• Should we keep the numpy style? The result of any operation follows the numpy rules.
• Or mark the difference to numpy in the docs and let the user know, like the .sum will return a scalar in some cases.