Confusing bare KeyError when LoRA config uses "alpha" instead of "scale"

[arjunneema09]

Description
The LoRA config format (lora_parameters in the YAML, e.g. mlx_lm/examples/lora_config.yaml)
requires a key named scale. Many LoRA tutorials and other frameworks (e.g. PEFT) use
alpha for the equivalent concept, so it's an easy mistake to make. When alpha is
passed instead of scale, the failure is a bare KeyError: 'scale' raised deep in
mlx_lm/tuner/utils.py (to_lora), with no message pointing at the actual problem.

To reproduce

1. Create a config with:

   lora_parameters:
     keys: ["self_attn.q_proj", "self_attn.v_proj"]
     rank: 8
     alpha: 16
     dropout: 0.0