Refactor BreedModel and tests

[russhwolf]

Things I'd like to change:

1. Currently BreedModel assumes callers will sequentially call refreshBreedsIfStale() and getBreedsFromCache(). We should have a single function that coordinates things internally so callers (both platform viewmodels and tests) don't have to.
2. Currently, getBreedsFromNetwork() returns a DataState, but this is actually only called internally inside refreshBreedsIfStale() which creates its own DataState. This gets a bit awkward because we need a when statement to convert the network call to the final state in refreshBreedsIfStale(), but not all cases of the when actually happen in practice (eg the loading state never triggers). We should wait to convert to a DataState until the last-mile function that gets called externally
3. getBreedsFromCache() will never emit if the database is empty. If we always refresh before calling it then this won't be a problem, but callers need to know this.
4. The naming and semantics of the test cases in BreedModelTest are inaccurate and confusing. staleDataCheckTest() asserts on errors that probably shouldn't be happening, and notifyErrorOnException() doesn't assert on any errors even though it expects them to happen ~5. KtorApiMock is currently a separate implementation of KtorApi. We could test our network/serialization logic more directly if it used DogApiImpl but passed a MockEngine to the httpClient.~ ~6. Should KtorApi be renamed DogApi?~

[russhwolf]

I did some partial work on this when starting #189, then pulled back because it was a bigger rabbit-hole than I realized. Was having trouble getting tests to pass with MockEngine and seeing some different behavior per-platform, so I'd like to do a bunch of cleanup and get this stuff all working better.

[kpgalligan]

Please review and see if still relevant. If not, close, otherwise, I guess let's chat or just do it if not huge?

[russhwolf]

Points 5 and 6 have been fixed but the rest of stuff in here is still outstanding