fix: improve performance in AI request logging

Instead of appending AI requests line by line to the output channel, we
now append them in a single step. Since appending to a channel is a
relatively expensive operation, this change significantly improves
performance for large requests.

This is particularly noticeable when AI inline code completions are
enabled for large files, as many requests may be sent during typing.

Author: sdirix

Diff for: packages/ai-code-completion/src/browser/code-completion-agent.ts

@@ -114,6 +114,9 @@ export class CodeCompletionAgentImpl implements CodeCompletionAgent {
             const requestId = generateUuid();
             const request: LanguageModelRequest = {
                 messages: [{ type: 'text', actor: 'user', query: prompt }],
+                settings: {
+                    stream: false
+                }
             };
             if (token.isCancellationRequested) {
                 return undefined;

Diff for: packages/ai-core/src/browser/frontend-language-model-registry.ts

@@ -366,7 +366,7 @@ const languageModelOutputHandler = (
                     'Sending request:'
                 );
                 const formattedRequest = formatJsonWithIndentation(args[0]);
-                formattedRequest.forEach(line => outputChannel.appendLine(line));
+                outputChannel.append(formattedRequest.join('\n'));
                 if (args[1]) {
                     args[1] = new Proxy(args[1], {
                         get<CK extends keyof CancellationToken>(

Diff for: packages/ai-openai/src/node/openai-language-model.ts

@@ -66,7 +66,7 @@ export class OpenAiModel implements LanguageModel {
         const settings = this.getSettings(request);
         const openai = this.initializeOpenAi();
 
-        if (this.isNonStreamingModel(this.model)) {
+        if (this.isNonStreamingModel(this.model) || (typeof settings.stream === 'boolean' && !settings.stream)) {
             return this.handleNonStreamingRequest(openai, request);
         }