WIP: Use dense LAPACK routines via block-tridiagonal view for banded * and qr

[dlfivefifty]

This should make QR and banded * banded operations much faster by leveraging fast dense linear algebra for blocks

[codecov[bot]]

Codecov Report

Merging #159 into master will decrease coverage by 1.16%. The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master     #159      +/-   ##
==========================================
- Coverage   72.77%   71.60%   -1.17%     
==========================================
  Files          21       22       +1     
  Lines        2516     2557      +41     
==========================================
  Hits         1831     1831              
- Misses        685      726      +41     

Impacted Files	Coverage Δ
src/generic/diagblockview.jl	0.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update da0c126...15addda. Read the comment docs.

[MasonProtter]

What's needed for this to merge? Just more tests?

[dlfivefifty]

It hasn't been done yet, it's just a bunch of experiments at the moment.

I got disheartened when I realized LAPACK's tpqrt2! is slower than just copying to a matrix and doing a dense QR.

For banded * banded, the issue is that there is apparently no BLAS triangular * triangular.