RetryingHttpClient.executeAsync skips retry handling when delegate throws synchronously

Description

RetryingHttpClient.executeAsync() calls the delegate before it has a CompletableFuture to attach its retry policy to:

val responseFuture = httpClient.executeAsync(requestWithRetryCount, requestOptions)

A custom HttpClient implementation is allowed by the SDK surface and can throw synchronously from executeAsync() instead of returning an exceptionally completed future.

When that happens, the exception escapes RetryingHttpClient.executeAsync() immediately. The existing handleAsync branch never sees it, so even retryable synchronous failures such as IOException, OpenAIIoException, or OpenAIRetryableException bypass the configured retry policy.

Expected behavior

Synchronous failures from the async delegate should enter the same retry decision path as exceptionally completed futures. Retryable failures should retry up to maxRetries; non-retryable failures should surface through the returned future.

Suggested fix

Wrap the delegate call and convert a synchronous throwable into an exceptionally completed CompletableFuture<HttpResponse> before applying the existing retry handler.

Add regression coverage for:

• a synchronous OpenAIRetryableException followed by a successful async response;
• a synchronous non-retryable exception completing the returned future exceptionally without a retry;
• retry-count headers remaining correct across the synchronous-failure retry.

Start at RetryingHttpClient.executeAsync() and inspect the existing handleAsync retry path around the delegate call. Run the current retry-related tests, then add regression coverage for synchronous retryable and non-retryable failures and verify retry-count headers remain correct; done means synchronous failures follow the same policy as exceptional futures.