Validate accuracy of standard rules in various domains

We should test our standard rules to confirm accuracy when working in various domains. This was started in #30, but there are still many improvements which should be made, both to the test suite and to the rules themselves.

- [ ] The variable substitutions provided by the test suite are very naive, substituting the same test value for all variables. We should test combinations of different variable types.
- [ ] We currently only test on integers, which should be changed. We should be testing using various common types (`Complex`, `Matrix`, etc.) and updating rule variable images accordingly.
- [ ] We currently don't test differentiation, since this is not a functionality that exists in `Base`, but we should probably test for various values using numerical approximations to validate the rules in `DiffRules.jl`.

Start with the work from #30 and inspect the test suite's variable substitutions and rule variable images. Compare the current integer-only coverage with Complex and Matrix cases, then assess numerical differentiation for DiffRules.jl in light of Base's limitations. Done means the standard rules are tested across varied variable combinations and common types, with differentiation coverage where applicable.

Written by the indexing model from the issue text.