Expose model capabilities (input/output types)

Currently, there is no way to check what input/output types a model supports (e.g. text, audio, image).

Problem

When building features like model selection UIs or validation, developers need to know if a model supports:

• text → text
• audio → text
• text → audio

Right now, this requires guessing based on model names (e.g. transcribe, audio, etc.), which is not reliable.

Expected Solution

Provide a way to access model capabilities, for example:

$model = AiClient::model('gpt-4o');

$model->getCapabilities();

/*
[
  'input' => ['text', 'image'],
  'output' => ['text'],
]
*/

Or simple helpers:

$model->supportsInput('text');
$model->supportsOutput('text');

Why this helps

• Makes model selection safer
• Removes fragile string-based checks
• Helps build better UIs and integrations

The issue names AiClient::model() and the proposed getCapabilities(), supportsInput(), and supportsOutput() entry points. Start by tracing how models are represented and how their input and output types could be exposed consistently. Done means callers can inspect text, audio, and image capabilities without relying on model-name guesses.