💪 General improvements to sentence generation

[beverage]

Now that highly configurable sentence generation generally works, and that the most basic of web services (and hosting) are up and running, it needs a lot of refinement so it can be consume by problem generators.

Mostly this is DTO-related, as almost all headaches right now are because of everything being json strings passing around everywhere. There are some bugs too, and some prompt refinement required as well.

The following minimally should be done:

• [x] ~Create DTOs for both input and output to OpenAI, in support of the next item.~
• [x] Update all prompts to properly use enum comparisons. Python truthiness is simply string existence, and testing explicit string comparisons. Everything until the last change was automatically true.
• [x] ~Update prompts to use said DTOs.~
• [x] Update random sentence generator to call explicit sentence generator and then remove it.
• [x] Update explicit sentence generator to accept all of its parameters.
• [ ] Add a 'correctness redrive' prompt, to validate sentence prompt results.

These together should create much better results for bundling into non-random problems.

[beverage]

Random sentence generator fix: f1a546054c21b478d1d66edbf1a5e84f60ecf3b9