Faster B[:,i] for a BandedMatrix using broadcasting

[jishnub]

Currently, broadcasting is faster than setindex! as SIMD gets blocked in the latter (JuliaLang/julia#40962 (comment)), so this change uses the faster variant. This might need revisiting in the future if setindex! becomes significantly faster.

On master

julia> B = brand(6000,6000,4000,4000);

julia> @btime $B[:, 3];
  4.985 μs (4 allocations: 47.00 KiB) # master
  3.980 μs (4 allocations: 47.00 KiB) # PR

julia> @btime $B[:, 5000];
  5.619 μs (4 allocations: 47.00 KiB) # master
  4.462 μs (4 allocations: 47.00 KiB) # PR

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (e445b49) 89.55% compared to head (33229d7) 89.55%.
Report is 11 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #421   +/-   ##
=======================================
  Coverage   89.55%   89.55%           
=======================================
  Files          25       25           
  Lines        3542     3544    +2     
=======================================
+ Hits         3172     3174    +2     
  Misses        370      370           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.