fix: replace AiGateway.embedBatch with direct LiteLLM cURL for upload indexing

AiGateway uses getenv(LITELLM_MASTER_KEY) + stream_context HTTP which was
failing on the chloe virtualhost process. New dbnToolsLiteLLMEmbedBatch()
helper mirrors dbnToolsCallGpuLlm — hardcoded URL + key, cURL-first, same
pattern already proven for LLM calls. Removes AiGateway dependency from
DeepResearchAgent entirely.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

includes/DeepResearchAgent.php

@@ -14,7 +14,6 @@ final class DbnDeepResearchAgent
     private const POOL_CAP = 30;
 
     private DbnAzureOpenAiGateway $azure;
-    private ?AiGateway $ai = null;
     private array $uploadVecs = [];
     private array $stepTimings = [];
 
@@ -50,9 +49,7 @@ final class DbnDeepResearchAgent
         dbnToolsBootCaveau();
         $aiPortalRoot = dbnToolsAiPortalRoot();
         require_once $aiPortalRoot . '/platform/includes/dbn_v6.php';
-        require_once $aiPortalRoot . '/lib/ai/AiGateway.php';
 
-        $this->ai = new AiGateway();
         $this->uploadVecs = [];
         $this->stepTimings = [];
 
@@ -141,7 +138,7 @@ final class DbnDeepResearchAgent
         if ($uploadChunks) {
             try {
                 $texts = array_map(fn(array $c) => $c['text'], $uploadChunks);
-                $vecs = $this->ai->embedBatch($texts, 'nomic-embed-text');
+                $vecs = dbnToolsLiteLLMEmbedBatch($texts);
                 if (count($vecs) === count($uploadChunks)) {
                     foreach ($uploadChunks as $i => $chunk) {
                         $this->uploadVecs[] = [
@@ -531,7 +528,7 @@ PROMPT;
             return [];
         }
         try {
-            $qVec = $this->ai->embed($question, 'nomic-embed-text');
+            $qVec = dbnToolsLiteLLMEmbedBatch([$question])[0] ?? [];
         } catch (Throwable $e) {
             error_log('DBN deep research sub-Q embed failed: ' . $e->getMessage());
             return [];