Adding default parameter for max_new_tokens in TransformersModel

[matfrei]

Related to issue #414.
This PR adds a default value of 5000 tokens for max_new_tokens amd max_tokens to TransformersModel.__call__ to avoid using the corresponding transformers default value (= 20 tokens) which is too low for agentic tasks.

Usage example:
Modify the last few lines of the Text2SQL example to be

from smolagents import CodeAgent, TransformersModel

agent = CodeAgent(
    tools=[sql_engine],
    model=TransformersModel("meta-llama/Meta-Llama-3.1-8B-Instruct"),
    max_steps=5
)

agent.run("Can you give me the name of the client who got the most expensive receipt?")

[aymeric-roucher]

I'd say it's better to set this ddefault max_new_tokens value directly in the init! The way parameters in models work is that any parameter passed in call will override the init value anyway.

[matfrei]

Yeah, I was kinda divided on that. I chose this solution as found this less likely to break unexpectedly should the interface ever change.
But sure, I'll move it to init!