New method for `sum` for `SubArray` of `SparseMatrix`

[btmit]

I often find myself needing to operate on a subset of a sparse array, typically a few adjacent columns. Creating a new SparseMatrix is relatively expensive, so being able to operate on view would be a big win.

Here's an example

using LinearAlgebra, SparseArrays, BenchmarkTools
s = sprand(101, 99, 0.01);

sc = s[:, 5:11];
sv = view(s, :, 5:11);

# sum parent array
@btime sum($s, dims=1);  # 400.706 ns (4 allocations: 1.02 KiB)
@btime sum($s, dims=2);  # 575.319 ns (9 allocations: 1.97 KiB)

# sum copy of columns
@btime sum($sc, dims=1);  # 244.282 ns (4 allocations: 256 bytes)
@btime sum($sc, dims=2);  # 454.545 ns (9 allocations: 1.97 KiB)

# sum view of columns
@btime sum($sv, dims=1);  # 1.179 μs (5 allocations: 416 bytes)
@btime sum($sv, dims=2);  # 6.442 μs (9 allocations: 1.88 KiB)

It looks like there is a specialized method for 1D column views. It would be great to extend this to groups of columns.