Faster indexing for BandedGeneralizedEigenvectors

[jishnub]

Similar to #354. This PR introduces an abstract type AbstractBandedEigenvectors and defines the indexing for that, so that its subtypes BandedEigenvectors and BandedGeneralizedEigenvectors may share the behavior. A BandedGeneralizedEigenvectors may reuse the scratch vector in the enclosed BandedEigenvectors, so no extra allocation is necessary.

After this,

julia> A = Symmetric(brand(Float64, 6, 6, 2, 4));

julia> B = Symmetric(BandedMatrix(0=>Float64.(1:6)));

julia> Λ, V = eigen(A, B);

julia> @btime $V[1,1];
  96.803 ns (0 allocations: 0 bytes)

whereas, on master,

julia> @btime $V[1,1];
  377.146 ns (1 allocation: 368 bytes)

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.70 🎉

Comparison is base (c32e3e0) 86.73% compared to head (d9ac8f7) 87.44%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #364      +/-   ##
==========================================
+ Coverage   86.73%   87.44%   +0.70%     
==========================================
  Files          24       22       -2     
  Lines        3378     2963     -415     
==========================================
- Hits         2930     2591     -339     
+ Misses        448      372      -76     

Impacted Files	Coverage Δ
src/symbanded/bandedeigen.jl	81.73% <100.00%> (+1.20%)	⬆️

... and 20 files with indirect coverage changes

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