Fixes #997

Problem

When LLM responses fail JSON parsing or schema validation, the error message only shows the parse error but not the raw response from the model. This makes it nearly impossible to debug malformed outputs without modifying the underlying model server code to capture responses separately.

Solution

Include the raw response content in the error message for generateObject and streamObject methods in the OpenAI and Ollama providers (Groq inherits from OpenAI, so it is covered too).

Before:

Error: Error parsing response from OpenAI: [{"expected":"object","code":"invalid_type",...}]

After:

Error: Error parsing response from OpenAI: [{"expected":"object","code":"invalid_type",...}]
Raw response: {"key": "unexpected_value", ...}

Testing

Manually verified the error message format change in both providers. No behavioral change — only the error message content is extended.

Summary by cubic

Adds the raw LLM response to JSON parse and schema validation error messages to make debugging malformed outputs easier. Applies to OpenAI and Ollama providers; Groq inherits the change. Fixes #997.

• Bug Fixes
  • OpenAI: append the raw message content in generateObject and the chunk text in streamObject when parsing fails.
  • Ollama: append the raw message content in generateObject when parsing fails.
  • No functional changes; only fuller error text.

^{Written for commit bcee37a. Summary will update on new commits.}