feat: model selection logic

[0xturboblitz]

• new configs: performance/cost priority, maxConcurrentCalls, addQuestions
• encapsulated model selection logic
• using tokenizer of right model token for estimation
• update input token price for GPT-3

[samheutmaker]

Hey this looks great. I'll review in the next couple days and get this merged.