Would you consider submitting it to the Function Repository?

For 'modify' part, don't you also need an inverse definitions to import an existing COMSOL expression into Mathematica for simplification?

I don't get your meaning of inverse definition, but you can access Comsol expressions by expanding the "equations" tag, where you can use Mathematica latter to parse the data by, for example, converting ewfd.Ez to a string "ewfd.Ez" or just stripping the ewfd tag. You can use Matlab to import all such equations, parse it into Mathematica, create a rule-based substitution and end-up with a simplified version of an equation.

That depends on who "we" is. Physicist usually call a two-index tensor a matrix, even if it is a 1 by 1 tensor. I know that many engineering books just fall back to "row matrices" and "column matrices" for vectors.

As a physicist myself, I'm not part of this "norm"... Tensors in physics have a physical meaning, matrices don't. Matrices are far too general, they don't need to transform covariantly. But call all you want, a rank 17 tensor in 1 dimension is still a scalar, no matter how pretty and more sophisticated it would be to be to call it a tensor. An equally valid point would be to call all real numbers Octonions with zero imaginary part, and argue that there are no "true" real numbers... But I don't think such pedantic matters on semantics has much to add to the discussion. One should use the lingo which one is used to...

Matlab indeed treats everything as a matrix (I never argued to oppose it), but not all definitions need to be "vectorized". You can be equally happy by making a function that only accepts scalars. As in Mathematica, not all functions have attribute Listable.

Matlab own error handling call 1x1 matrices scalars, so, why can't we all? Just because internally they are two-dimensional arrays? Should we still can a matrix a matrix if it has rank different then 2? Or call it an array? Tensor?

In Comsol, you can only modify vectors, tensors, component-wisely, so, there is no need to have a high-level representation in terms of dot products, matrix multiplication etc.