Add reasoning events and abort-aware finish reason to chat endpoint

Map thinking_start/delta/end to Vercel AI SDK reasoning-start/delta/end
chunk types. Derive finish reason from enqueueMessage result - aborted
sessions get a clean finish with reason "error" instead of an error chunk.

packages/coding-agent/src/core/gateway-runtime.ts

@@ -638,7 +638,12 @@ export class GatewayRuntime {
 				if (result.ok) {
 					finishVercelStream(response, "stop");
 				} else {
-					errorVercelStream(response, result.error ?? "Unknown error");
+					const isAbort = result.error?.includes("aborted");
+					if (isAbort) {
+						finishVercelStream(response, "error");
+					} else {
+						errorVercelStream(response, result.error ?? "Unknown error");
+					}
 				}
 			}
 		} catch (error) {

packages/coding-agent/src/core/vercel-ai-stream.ts

@@ -125,6 +125,25 @@ export function createVercelStreamListener(
 							input: inner.toolCall.arguments,
 						});
 						return;
+					case "thinking_start":
+						writeChunk(response, {
+							type: "reasoning-start",
+							id: `reasoning_${inner.contentIndex}`,
+						});
+						return;
+					case "thinking_delta":
+						writeChunk(response, {
+							type: "reasoning-delta",
+							id: `reasoning_${inner.contentIndex}`,
+							delta: inner.delta,
+						});
+						return;
+					case "thinking_end":
+						writeChunk(response, {
+							type: "reasoning-end",
+							id: `reasoning_${inner.contentIndex}`,
+						});
+						return;
 				}
 				return;
 			}