Remove LLM_SETTINGS

[LOCKhart07]

Let's get this info from the counter_callback and delete the superfluous LLM_SETTINGS. Having the same info in two places is not a good practice. This will also help in the refactoring related to valory-xyz#319 later on. Here's an example:

mech-predict/packages/valory/customs/prediction_request_lite/prediction_request.py

Lines 783 to 788 in 511612b

	token_prices = getattr(counter_callback, TOKEN_COSTS_PER_MODEL_ATTR, {})
	tokens_config = token_prices.get(engine, {})
	if not tokens_config:
	raise ValueError("The tool cannot run without models' configurations.")
	default_max_tokens = tokens_config.get(MAX_TOKENS)

Originally posted by @Adamantios in #65 (comment)