Compare libraries when several sequences are generated

[rlouf]

We currently only compare libraries for a single generation. However, the main advantage of Outlines is that generation time becomes negligible once the index has been compiled. We should also compare timings for 100 generations for each pattern.

To make sure that the conditions are as close as possible to actually generating text we should generate fake entries using the faker library. Failure to do so would overlook lm-format-enforcer's main downside.