Refactor Sequences and Generator pipeline into single LLM pipeline

[davidmezzetti]

Currently, the Sequences and Generator pipelines load sequence to sequence and text generation models respectively. Both of these types of models can be used in a generative way and the logic should be refactored to make txtai easier to use with LLMs.

Logic to determine if a model is a sequence-sequence or text generation model should be added. The Sequences and Generator pipeline should extend the LLM pipeline and be a simple way to force the model type.