chore(vendor): import pi 1.3.0

Imported as a linear vendor commit so forge rebases do not replay the upstream root tree.

git-subtree-dir: third_party/pi

git-subtree-split: 71023aec1546d6d677a65669c5665c0927ce0dd0

third_party/pi/packages/ai/test/context-estimate.test.ts

@@ -1,7 +1,8 @@
 import { describe, expect, it } from "vitest";
 import { buildBaseOptions } from "../src/api/simple-options.ts";
-import type { AssistantMessage, Context, Model, Usage } from "../src/types.ts";
+import type { AssistantMessage, Model, Usage } from "../src/types.ts";
 import { estimateContextTokens } from "../src/utils/estimate.ts";
+import { normalizeContext } from "../src/utils/transcript.ts";
 
 function createUsage(totalTokens: number): Usage {
 	return {
@@ -41,27 +42,42 @@ const model: Model<"openai-responses"> = {
 };
 
 describe("context token estimation", () => {
+	// Regression for #10497: large new inputs need more room than chars/4 allows.
+	it("reserves 3.5 characters per token for new text when limiting output", () => {
+		const context = normalizeContext({
+			messages: [createAssistant(100, 2_000), { role: "user", content: "x".repeat(3_500), timestamp: 200 }],
+		});
+
+		expect(estimateContextTokens(context)).toEqual({
+			tokens: 3_000,
+			usageTokens: 2_000,
+			trailingTokens: 1_000,
+			lastUsageIndex: 0,
+		});
+		expect(buildBaseOptions(model, context).maxTokens).toBe(2_904);
+	});
+
 	it("ignores stale assistant usage after a newer message is inserted before it", () => {
-		const context: Context = {
+		const context = normalizeContext({
 			systemPrompt: "system",
 			messages: [
 				{ role: "user", content: "summary", timestamp: 200 },
 				createAssistant(100, 9_500),
 				{ role: "user", content: "x".repeat(4_000), timestamp: 300 },
 			],
-		};
+		});
 
 		expect(estimateContextTokens(context)).toEqual({
-			tokens: 1_005,
+			tokens: 1_149,
 			usageTokens: 0,
-			trailingTokens: 1_005,
+			trailingTokens: 1_149,
 			lastUsageIndex: null,
 		});
-		expect(buildBaseOptions(model, context).maxTokens).toBe(4_899);
+		expect(buildBaseOptions(model, context).maxTokens).toBe(4_755);
 	});
 
 	it("uses assistant usage again after a response to the inserted context", () => {
-		const context: Context = {
+		const context = normalizeContext({
 			messages: [
 				{ role: "user", content: "summary", timestamp: 200 },
 				createAssistant(100, 9_500),
@@ -69,12 +85,12 @@ describe("context token estimation", () => {
 				createAssistant(400, 2_000),
 				{ role: "user", content: "tail", timestamp: 500 },
 			],
-		};
+		});
 
 		expect(estimateContextTokens(context)).toEqual({
-			tokens: 2_001,
+			tokens: 2_002,
 			usageTokens: 2_000,
-			trailingTokens: 1,
+			trailingTokens: 2,
 			lastUsageIndex: 3,
 		});
 	});