Compatibility with backslash and sparse arrays

So I am not sure what the issue is, or if I am not doing things properly, but it seems that it should be doable to make `Unitful.jl` work with linear algebra. I tend to do array multiplications and solve linear systems a lot via backslash. Here is an example of unit-less things I do:
```julia
# no-unit Array example that works
κ = rand(3, 3)
c = rand(3)
s = κ * c
c ≈ κ \ s

# no-unit sparse Array example that works
using SparseArrays, LinearAlgebra
κ = sprand(3, 3, 0.5) + I
c = rand(3)
s = κ * c
c ≈ κ \ s
```

Now doing that with unit-full scalars is not an issue:
```julia
using Unitful
# Scalar example that works
κ = rand()u"1/s"
c = rand()u"mol/m^3"
s = κ * c
c ≈ κ \ s
```

But it breaks on two important things for me:
1. Using `\` on arrays raises an error:
```julia
# Array example that works for multiplication only
κ = rand(3, 3)u"1/s"
c = rand(3 )u"mol/m^3"
s = κ * c
c ≈ κ \ s # Gives DimensionError
```
My guess is that the factorization is a problem, as I do not know how would one split the units of `κ` into the units of its factors. (E.g., what are the units of `L` and `U` in `κ = L * U`?). Is it possible to give a single unit to a whole array and tell `\` to do the unit conversion on the side? (I.e., do the `κ \ s` evaluation *without* units, and assign it the unit of `s` divided by the unit of `κ`.)

2. Using `*` on sparse arrays raises an error:
```julia
# Sparse Array example that does not work
κ = (sprand(3, 3, 0.5) + I)u"1/s"
c = rand(3)u"mol/m^3"
s = κ * c # Gives DimensionError
```
My guess here is it comes from the zeros of the sparse matrix, which are not given a unit? Maybe doing the unit conversion on the side is still doable?

**EDIT**: I realize this is a partial duplicate of issue #46. The only new feature request here is thus the sparse matrix one.