Improve complexity of getindex

[tpgillam]

Previously, getindex constructed the full dense matrix. This gets expensive, as it is used when e.g. forming a string representation of the array.

Now we compute only the part that is necessary.

Closes #30

[codecov]

Codecov Report

Merging #31 (74e5945) into master (f0f1cdf) will increase coverage by 1.36%.
The diff coverage is 90.90%.

@@            Coverage Diff             @@
##           master      #31      +/-   ##
==========================================
+ Coverage   71.84%   73.21%   +1.36%     
==========================================
  Files           4        3       -1     
  Lines         103      112       +9     
==========================================
+ Hits           74       82       +8     
- Misses         29       30       +1     

Impacted Files	Coverage Δ
src/woodbury_pd_mat.jl	93.47% <90.90%> (-0.97%)	⬇️
src/PDMatsExtras.jl

---

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 f0f1cdf...74e5945. Read the comment docs.

[mzgubic]

Just to understand: previously the string would iterate over all the indexes, which would each construct a full Matrix, and then return the value. Now, instead of constructing the full matrix, we just construct the right element.

This seems sensible, and I wonder whether it takes roughly the same time as string(Matrix(woodbury))?