My understanding is that Mathworks is moving toward a more object-oriented approach for a lot of the Statistics Toolbox functions.
stepwiselm() is one of the newer functions, and it returns a linear model object, which can then be used with various methods to extract info from the model.
stepwisefit is one of the older function. It directly returns output (e.g. model coefficients) into individual variables.
Algorithmically, I don't expect there is any difference, but I cannot say that definitively.
