Support for single indexes in MatrixExpr

[jcrist]

This allows MatrixSymbols to be iterated over, and gives a similar
interface to DenseMatrix. MatExpr are treated as row-major order arrays.
This only works if both the matrix shape and index are Integer/int.

Demo:

>>> qs = symbols('q0:5')
>>> q = MatrixSymbol('q', 5, 1)
>>> expr = sum(qs); expr
q0 + q1 + q2 + q3 + q4

# Can be used to create a sub_dict, with all symbols in a vector representation.
# This is useful for code-generation or numerical evaluation
>>> sub_dict = dict(zip(qs, q)); sub_dict
{q4: q[4, 0], q3: q[3, 0], q0: q[0, 0], q1: q[1, 0], q2: q[2, 0]}
>>> expr.subs(sub_dict)
q[0, 0] + q[1, 0] + q[2, 0] + q[3, 0] + q[4, 0]

# Can also iterate over slices. This will sum just the fourth row of matrix A
>>> A = MatrixSymbol('A', 5, 5)
>>> sum(A[3, :])
A[3, 0] + A[3, 1] + A[3, 2] + A[3, 3] + A[3, 4]

[jcrist]

@mrocklin : Please review.

[jcrist]

Ping for review: @mrocklin.

[jcrist]

Can I get a review on this from someone? This will really help make codegen for mechanics stuff alot easier to express using vectors. @sympy/mechanics

[moorepants]

This seems fine. +1

I can't see why this would cause any issues. It is just a simple feature addition.

[mrocklin]

Sorry for being absent here. We had this at one point and I took it out. Perhaps it was because it was causing infinite loops on symbolic shapes. I see that you've handled that. No clear objection from me. I haven't inspected this that deeply though.

[jcrist]

No worries @mrocklin. We could make it work with symbolics by changine the indices to be symbolic expressions between the dimensions and the single index, but I don't really see the use of that (unless someone wanted to check C array index arithmetic). Thanks for the review. Merging.