Accept both text_completion and chat.completion in completions response

[bledden]

Summary

The completions.generate() method was failing with a validation error because the SDK expected object: "text_completion" but the OpenRouter API returns object: "chat.completion".

This caused the error:

ResponseValidationError: Response validation failed
ZodError: Invalid input: expected "text_completion"

Changes

• Updated CompletionResponse type to accept either "text_completion" or "chat.completion"
• Updated the Zod schema to use z.enum() instead of z.literal()
• Updated the OpenAPI spec to match
• Added E2E test for completions

Testing

const response = await client.completions.generate({
  model: 'meta-llama/llama-3.1-405b-instruct',
  prompt: '5 + 7 = ',
  maxTokens: 10,
  stream: false,
});
// Now works without validation errors

Fixes #101

[bledden]

Note: This fix allows the SDK to accept both text_completion and chat.completion as valid values for the object field in completions responses.

However, the underlying issue may be on the API side - the /api/v1/completions endpoint is returning object: "chat.completion" when it arguably should return "text_completion" to match OpenAI's completions API behavior.

The OpenRouter team may want to investigate whether the API response should be updated to return the correct object type for completions, or whether returning chat.completion is intentional behavior.