Some Adjoint and Transpose stuff

[willtebbutt]

This PR pretty much brings us up to feature-parity with Zygote on the rules for Adjoint and Transpose.

TODO:

• Enable the use of Composites in the reverse-pass
• Make the various transpose-like functions play nicely with complex numbers.
• Make parent return a Composite, rather than the transposition of ȳ, since ȳ might itself be e.g. a Composite.
• Test parent.

edit: the second and third points weren't achievable in this PR because our testing infrastructure isn't up to scratch at the minute. Opened this issue regarding the problem.

[codecov-io]

Codecov Report

Merging #324 (17b1350) into master (22eddb4) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #324   +/-   ##
=======================================
  Coverage   97.62%   97.62%           
=======================================
  Files          18       18           
  Lines        1011     1011           
=======================================
  Hits          987      987           
  Misses         24       24           

Impacted Files	Coverage Δ
src/rulesets/LinearAlgebra/structured.jl	98.90% <100.00%> (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 22eddb4...17b1350. Read the comment docs.

[sethaxen]

Overall, this looks right. It looks like this is causing some relevant integration errors with Zygote. Once those are fixed (or if turns out to be a Zygote bug), feel free to bump the version number and merge.