WIP: BandedBlockBandedMatrix can parameterize storage

[mdavezac]

But no tests for anything but PseudoBlockMatrix yet.

[mdavezac]

@dlfivefifty, the parameterization is pretty much done. I'll rebase on master once master passes on Travis. It can be tested by susbtituting PseudoBlockArray with BlockArray.

I noticed that "test_broadcasting.jl" is not in "runtests.jl". Just putting it out there if it got lost at some point.

[dlfivefifty]

I noticed that "test_broadcasting.jl" is not in "runtests.jl". Just putting it out there if it got lost at some point.

Whoops, that’s an oversight

[dlfivefifty]

I merged the changes to master back here, and tests should pass when on BlockArrays and BandedMatrices master.

[dlfivefifty]

I've added an example using SharedArrays for building matrices. (CLArrays.jl hasn't been updated to 1.0 and I don't have access to a CUDA device.) The current SharedArrays code is much slower than the Serial one, but this is perhaps because I'm not specifying localindices.

[dlfivefifty]

Now that I've merged (and squashed) the changes introduced here, this PR probably needs to be rebased.