fix(openai): only include parallel_tool_calls when explicitly enabled

This change reverts the parallel_tool_calls handling to only include it
when explicitly set to true (metadata?.parallelToolCalls === true).

Root cause: Commit ed35b09aa changed the default from false to true,
which caused models like GLM4.5 that do not support this parameter
to get stuck in a loop when reading files repeatedly.

The fix affects:
- src/api/providers/openai.ts (OpenAI-compatible handler)
- src/api/providers/base-openai-compatible-provider.ts (base class for other providers)

This ensures compatibility with models that do not support the
parallel_tool_calls parameter while still allowing it for models
that explicitly enable it.

Fixes #11071

src/api/providers/__tests__/openai.spec.ts

@@ -635,7 +635,6 @@ describe("OpenAiHandler", () => {
 					temperature: 0,
 					tools: undefined,
 					tool_choice: undefined,
-					parallel_tool_calls: true,
 				},
 				{ path: "/models/chat/completions" },
 			)
@@ -684,7 +683,6 @@ describe("OpenAiHandler", () => {
 					],
 					tools: undefined,
 					tool_choice: undefined,
-					parallel_tool_calls: true,
 				},
 				{ path: "/models/chat/completions" },
 			)

src/api/providers/base-openai-compatible-provider.ts

@@ -95,7 +95,7 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
 			stream_options: { include_usage: true },
 			tools: this.convertToolsForOpenAI(metadata?.tools),
 			tool_choice: metadata?.tool_choice,
-			parallel_tool_calls: metadata?.parallelToolCalls ?? true,
+			...(metadata?.parallelToolCalls === true && { parallel_tool_calls: true }),
 		}
 
 		// Add thinking parameter if reasoning is enabled and model supports it

src/api/providers/openai.ts

@@ -161,7 +161,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				...(reasoning && reasoning),
 				tools: this.convertToolsForOpenAI(metadata?.tools),
 				tool_choice: metadata?.tool_choice,
-				parallel_tool_calls: metadata?.parallelToolCalls ?? true,
+				...(metadata?.parallelToolCalls === true && { parallel_tool_calls: true }),
 			}
 
 			// Add max_tokens if needed
@@ -229,7 +229,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				// Tools are always present (minimum ALWAYS_AVAILABLE_TOOLS)
 				tools: this.convertToolsForOpenAI(metadata?.tools),
 				tool_choice: metadata?.tool_choice,
-				parallel_tool_calls: metadata?.parallelToolCalls ?? true,
+				...(metadata?.parallelToolCalls === true && { parallel_tool_calls: true }),
 			}
 
 			// Add max_tokens if needed
@@ -348,7 +348,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				// Tools are always present (minimum ALWAYS_AVAILABLE_TOOLS)
 				tools: this.convertToolsForOpenAI(metadata?.tools),
 				tool_choice: metadata?.tool_choice,
-				parallel_tool_calls: metadata?.parallelToolCalls ?? true,
+				...(metadata?.parallelToolCalls === true && { parallel_tool_calls: true }),
 			}
 
 			// O3 family models do not support the deprecated max_tokens parameter
@@ -382,7 +382,7 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				// Tools are always present (minimum ALWAYS_AVAILABLE_TOOLS)
 				tools: this.convertToolsForOpenAI(metadata?.tools),
 				tool_choice: metadata?.tool_choice,
-				parallel_tool_calls: metadata?.parallelToolCalls ?? true,
+				...(metadata?.parallelToolCalls === true && { parallel_tool_calls: true }),
 			}
 
 			// O3 family models do not support the deprecated max_tokens parameter