Fix `TGI` (Text Generation Inference) Endpoint Inference and TGI JSON Grammar Generation (#502)
* fix: Lighteval communication with TGI
* fix: JSON grammar constrained generation
* fix: unit tests +
add: dep in extra
* fix: request var => doc var after refactor
* fix: update test to support the new grammar field
* fix: TGI endpoint with the new refactor
* update: TGI model config in examples with the latest parameters
* add: example custom task on a classification dataset to demonstrate the usage of constrained grammar generation using TGI
* add: format example task
* fix: unit test
* add: adapt the in the yaml config to use similarly to the other endpoints
* clean: moved new task to community_tasks
* fix: format
* clean: delete unused grammar field
* del: grammar
* add: copyright at the top
* del: langcodes dep isn't needed anymore
* add: use load from file directly in the main endpoint
* del: newlines
* add: mock HTTP request for info to TGI server
---------
Co-authored-by: Nathan Habib <30601243+NathanHB@users.noreply.github.com>