Use outlines.processors for models.llamacpp

[lapp0]

Fixes #965

Problem

in main

• generate.fsm only supports SequenceGenerator,
• models.llamacpp doesn't support SequenceGenerator, it only supports SequenceGeneratorAdapter
• integrations.llamacpp doesn't have a FSM logits processor.

Solution

• Ensure models.llamacpp uses outlines.processors / SequenceGeneratorAdapter for all generators
• Update generate.fsm to use SequenceGeneratorAdapter for all unified models
• test_generate.py tests for generate.fsm on MLXLM, llamacpp, and transformers (the three models using outlines.processors)
  • fix critical generate.fsm bug discovered through this test in guide.py impacting all models