fix: preserve empty reasoning_content for DeepSeek V4 in non-streaming and streaming paths

Three truthy checks were dropping empty reasoning_content ('') from DeepSeek
V4's thinking mode responses, causing 'must be passed back to the API' errors
in multi-turn tool call chains:

1. Non-streaming parser: if (reasoning != null && reasoning.length > 0)
   -> Now: if (reasoning != null) to preserve empty strings
2. Streaming parser: if (reasoningContent) truthy check
   -> Now: if ('reasoning_text' in delta) to detect field presence
3. Outbound converter: if (part.text) reasoningText = part.text
   -> Now: reasoningText = part.text ?? '' to preserve empty text

PR anomalyco#24146 fixed the transform.ts path; this completes the remaining cases.

Author: ihoooohi

packages/opencode/src/provider/sdk/copilot/chat/convert-to-openai-compatible-chat-messages.ts

@@ -95,7 +95,9 @@ export function convertToOpenAICompatibleChatMessages(prompt: LanguageModelV3Pro
               break
             }
             case "reasoning": {
-              if (part.text) reasoningText = part.text
+              // Preserve empty reasoning text — some providers (e.g. DeepSeek V4)
+              // require reasoning_content: "" to be sent back in multi-turn chains.
+              reasoningText = part.text ?? ""
               break
             }
             case "tool-call": {

packages/opencode/src/provider/sdk/copilot/chat/openai-compatible-chat-language-model.ts

@@ -227,8 +227,10 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
     }
 
     // reasoning content (Copilot uses reasoning_text):
+    // Preserve empty reasoning_text (e.g. DeepSeek V4 returns reasoning_content: "")
+    // because some providers require it to be sent back verbatim in multi-turn chains.
     const reasoning = choice.message.reasoning_text
-    if (reasoning != null && reasoning.length > 0) {
+    if (reasoning != null) {
       content.push({
         type: "reasoning",
         text: reasoning,
@@ -477,22 +479,24 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
               reasoningOpaque = delta.reasoning_opaque
             }
 
-            // enqueue reasoning before text deltas (Copilot uses reasoning_text):
+            // enqueue reasoning before text deltas (Copilot uses reasoning_text).
+            // Handle empty reasoning_text (e.g. DeepSeek V4 may stream reasoning_content: "").
             const reasoningContent = delta.reasoning_text
-            if (reasoningContent) {
+            if ("reasoning_text" in delta) {
               if (!isActiveReasoning) {
                 controller.enqueue({
                   type: "reasoning-start",
                   id: "reasoning-0",
                 })
                 isActiveReasoning = true
               }
-
-              controller.enqueue({
-                type: "reasoning-delta",
-                id: "reasoning-0",
-                delta: reasoningContent,
-              })
+              if (reasoningContent) {
+                controller.enqueue({
+                  type: "reasoning-delta",
+                  id: "reasoning-0",
+                  delta: reasoningContent,
+                })
+              }
             }
 
             if (delta.content) {