Non-scalar Matrix Field bugs

[dennisYatunin]

Describe the bug

In the process of working the RosSSP timestepper, @Sbozzolo has uncovered 4 bugs in the Field/FieldMatrix broadcasting machinery:

• Division (not multiplication) of a Field of BandMatrixRows of non-scalars by a scalar throws an error [fixed in Remove unnecessary method specializations for Numbers #2014]
• Multiplciation/division of a FieldMatrix by a scalar throws an error
• Base.one of a FieldMatrix with non-scalar subblocks generates a FieldMatrix with scalar subblocks, which then throws an error when combined with a FieldMatrix similar to the original
• LinearAlgebra.ldiv! generates NaNs when using a FieldMatrix with a single UniformScaling block.

Steps to Reproduce

Using W to denote the ImplicitEquationJacobian used in ClimaAtmos, the bugs can be reproduced as follows:

• For the first bug, running W.matrix[@name(f.u₃), @name(f.u₃)] ./ 2 reproduces the error
• For the second bug, running W.matrix .* 2 reproduces the error
• For the third bug, running one(W.matrix) generates the incorrect result, and running W.matrix .+ one(W.matrix) reproduces the error
• For the fourth bug, ldiv!(x, W.matrix, b) with a matrix that contains a single UniformScaling block and a FieldVector x that is full of ones generates the incorrect result