fix: use max_completion_tokens instead of max_tokens for OpenAI-compatible APIs

Azure OpenAI and newer OpenAI models (o1, o3, o4...) reject `max_tokens`
with a 400 error and require `max_completion_tokens` instead.

Maps `params.max_tokens` → `max_completion_tokens` in the request body,
which is the current standard across OpenAI-compatible providers.

src/services/api/openaiShim.ts

@@ -636,9 +636,13 @@ class OpenAIShimMessages {
     const body: Record<string, unknown> = {
       model: request.resolvedModel,
       messages: openaiMessages,
-      max_tokens: params.max_tokens,
       stream: params.stream ?? false,
     }
+    if (params.max_tokens !== undefined) {
+      body.max_completion_tokens = params.max_tokens
+    } else if ((params as Record<string, unknown>).max_completion_tokens !== undefined) {
+      body.max_completion_tokens = (params as Record<string, unknown>).max_completion_tokens
+    }
 
     if (params.stream) {
       body.stream_options = { include_usage: true }