Support block/proc for dynamic model selection in Agent

[kryzhovnik]

Feature Request

Allow the model class method in RubyLLM::Agent to accept a block for dynamic model selection at runtime, consistent with how tools, params, headers, and schema already support blocks.

Use case

Selecting a model based on runtime context — e.g., routing simple tasks to a cheaper model:

class CardAgent < RubyLLM::Agent
  inputs :card
  model { card.special_type? ? 'gpt-4.1-mini' : 'gpt-4.1-nano' }
  instructions 'You are helpful.'
end

CardAgent.chat(card: @card)

Current behavior

model only accepts a string:

model 'gpt-4.1-nano'

Proposed behavior

model also accepts a block, evaluated in the same runtime context as other block-based DSLs (tools, params, etc.), with access to declared inputs:

inputs :quality
model { quality == :high ? 'gpt-4.1-mini' : 'gpt-4.1-nano' }

Implementation

I have a working implementation ready (~13 lines of production code) that follows the existing apply_* pattern. Happy to open a PR if this aligns with the project's direction.