The signature of Matrix::map() seems unnecessarily restrictive

Currently, Matrix::map() is defined as follows:

fn map N>(&self, f: F) -> Matrix

The restriction that the mapping function's return type must be the same as its input type, which does not exist when using std iterators for example, forbids several kind of useful mappings, including:

- Mapping between different scalar types (e.g. i32 <-> f32 <-> f64 <-> num::Complex)
- Partial reductions (e.g. turning a vector of vectors into a vector of scalars by folding the inner vectors)

This can be hacked around through a combination of iter() and FromIterator, but the resulting code will be relatively verbose.

Is there any particular reason why this is the case?