Check isbanded in conversions

[jishnub]

While converting from an AbstractMatrix to a BandedMatrix, we may use a faster path if the argument is banded and has a BandedColumns{DenseColumnMajor}() memory layout .
On master

julia> D = Diagonal(rand(1000));

julia> @btime convert(BandedMatrix, $D);
  4.521 μs (4 allocations: 8.11 KiB)

julia> @btime BandedMatrix{Float32}($D);
  4.588 μs (2 allocations: 4.11 KiB)

julia> @btime BandedMatrix($D);
  2.396 μs (2 allocations: 7.98 KiB)

This PR

julia> @btime convert(BandedMatrix, $D);
  759.925 ns (6 allocations: 8.20 KiB)

julia> @btime BandedMatrix{Float32}($D);
  564.160 ns (4 allocations: 4.20 KiB)

julia> @btime BandedMatrix($D);
  641.483 ns (4 allocations: 8.06 KiB)

With some refactoring, all of these could become equally performant.

TBH I'm a bit surprised that this is necessary at all, since I would have thought that banded broadcast already handles this. Presumably, much of the gain comes from linear indexing and vectorization.

[codecov]

Codecov Report

Base: 80.47% // Head: 80.59% // Increases project coverage by +0.12% 🎉

Coverage data is based on head (f357dfa) compared to base (7315691).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #289      +/-   ##
==========================================
+ Coverage   80.47%   80.59%   +0.12%     
==========================================
  Files          22       23       +1     
  Lines        3262     3283      +21     
==========================================
+ Hits         2625     2646      +21     
  Misses        637      637              

Impacted Files	Coverage Δ
src/banded/BandedMatrix.jl	84.68% <100.00%> (+0.55%)	⬆️
src/interfaceimpl.jl	100.00% <100.00%> (ø)
src/BandedMatrices.jl	100.00% <0.00%> (ø)
src/generic/Band.jl	71.60% <0.00%> (+0.35%)	⬆️
src/lapack.jl	86.36% <0.00%> (+0.48%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.