allow generators.Base.generate() to take an optional param specifying generation count

[leondz]

Also required clearing up some type signatures and amending _call_model a bunch of places

Would rather _call_model didn't have to take an extra param, but sometimes it cares about the number of generations. Because generator objects handle parallel requests, they're not a great place to handle state - each generation request should be atomic and self-contained

Leaving in _call_model's generations_this_call param makes for a weird situation where one might think _call_model can be invokes asking for many generations, where in fact this may be ignored (undef behaviour)

[leondz]

oh hey we should be checking if models generate the number of responses we want, and define & test what happens when they don't. cool cool cool (#601 )

[erickgalinkin]

Looks good to me!