Improve compatibility with OpenRouter (DecodingError in CompletionTokensDetails)

[ruslanhirychau]

Hi Kevin!

First of all, thanks for the great library! It's very clean and easy to use with OpenAI directly.

I’m currently using it with OpenRouter as an endpoint, and I've encountered a small issue when streaming responses. It seems like OpenRouter (and some other providers) doesn't always include all the fields in the usage object that the standard OpenAI API does.

I used Gemini to help me debug the issue. It identified that the DecodingError (keyNotFound) for accepted_prediction_tokens occurs because the default Swift decoder expects the keys to be present in the JSON if they are defined in CodingKeys, even if the properties are optional.

I’ve created a fork, applied the suggested fix, and confirmed that it works perfectly with OpenRouter.

The Suggested Fix:
Implement a custom init(from decoder: Decoder) using decodeIfPresent for CompletionTokensDetails to make it more robust:

public struct CompletionTokensDetails: Decodable, Sendable {
    public let acceptedPredictionTokens: Int?
    public let rejectedPredictionTokens: Int?
    public let reasoningTokens: Int?
    
    private enum CodingKeys: String, CodingKey {
        case acceptedPredictionTokens = "accepted_prediction_tokens"
        case reasoningTokens = "reasoning_tokens"
        case rejectedPredictionTokens = "rejected_prediction_tokens"
    }

    public init(from decoder: Decoder) throws {
        let container = try decoder.container(keyedBy: CodingKeys.self)
        self.acceptedPredictionTokens = try container.decodeIfPresent(Int.self, forKey: .acceptedPredictionTokens)
        self.rejectedPredictionTokens = try container.decodeIfPresent(Int.self, forKey: .rejectedPredictionTokens)
        self.reasoningTokens = try container.decodeIfPresent(Int.self, forKey: .reasoningTokens)
    }
}

Could you please consider incorporating this change (or a similar fix) into your library to improve compatibility with alternative providers?

Keep up the great work! Best regards.