Allow providing options in alias or model definition

It would be very convenient to allow defining an alias such as o3-mini-high that is an alias for o3-mini but also automatically supplies -o reasoning high. In particular, it makes it easier to use shell completions and to swap out one model for the other when testing a different response to the previously executed llm shell command from bash/shell history (only edit the -m payload and not have to also add or remove 3 additional arguments, especially if you're comparing, say, o3-mini-medium vs gemini-2.5 vs o3-mini-high or whatever where it's not the same base model).

Alternatively, if it could be defined in a models .yaml instead of in the alias.yaml as a separate model, for example

- model_id: o3-mini-high
  model_name: o3-mini
  model_args:
  - reasoning: high

- model_id: o3-mini-medium
  model_name: o3-mini
  model_args:
  - reasoning: medium

Start by tracing how alias.yaml and model YAML definitions are loaded for the llm shell command, then compare the two proposed configuration forms. Done means an alias or model definition can map to a base model while automatically supplying reasoning options, allowing model swaps by changing only the -m value.