fix(ai/recipes): declare max_batch_tokens on google embedding recipe

src/core/ai/recipes/google.ts

@@ -14,6 +14,13 @@ export const google: Recipe = {
       models: ['gemini-embedding-001'],
       default_dims: 768,
       dims_options: [768, 1536, 3072],
+      // Per-text cap is 2048 tokens; per-request total ~20k tokens before
+      // gemini-embedding-001 starts 429-ing. chars_per_token: 2 covers
+      // CJK-dense corpora (English averages ~4, CJK ~1.5); safety_factor
+      // stays at the gateway default 0.8 so pre-split lands at ~8 000
+      // chars/batch — well under any per-request floor Google publishes.
+      max_batch_tokens: 20_000,
+      chars_per_token: 2,
       cost_per_1m_tokens_usd: 0.15,
       price_last_verified: '2026-04-20',
     },

test/ai/adaptive-embed-batch.test.ts

@@ -375,16 +375,22 @@ describe('startup warning for recipes missing max_batch_tokens', () => {
       console.warn = original;
     }
 
-    // The warning text should match the documented contract.
+    // The warning text contract is still documented; after v0.34.5 (google
+    // declared max_batch_tokens) every first-party native/openai-compat
+    // recipe is capped, so the canary set is empty. The mechanism itself
+    // is still exercised by the once-per-process suppression check above
+    // (firstCallCount stability across re-configure).
     const contractMatch = warnings.filter(w =>
       w.includes('[ai.gateway]') && w.includes('declares an embedding touchpoint'),
     );
-    expect(contractMatch.length).toBeGreaterThan(0);
+    expect(contractMatch.length).toBe(0);
 
     // Voyage declares max_batch_tokens → suppressed. OpenAI is the
     // canonical fast-path recipe → also suppressed by id. Both must be
     // absent from the warnings.
     expect(warnings.find(w => w.includes('"voyage"'))).toBeUndefined();
     expect(warnings.find(w => w.includes('"openai"'))).toBeUndefined();
+    // After v0.34.5 google also declares max_batch_tokens.
+    expect(warnings.find(w => w.includes('"google"'))).toBeUndefined();
   });
 });

test/ai/no-batch-cap-suppression.serial.test.ts

@@ -52,15 +52,20 @@ describe('v0.32 #779: no_batch_cap suppresses the missing-max_batch_tokens warni
     }
   });
 
-  test('configureGateway STILL warns for google (real provider, no cap declared)', () => {
+  test('configureGateway does not warn for any first-party recipe (every embedding recipe declares max_batch_tokens or no_batch_cap)', () => {
     warnSpy.mockClear();
     resetGateway();
     configureGateway({ env: {} });
     const messages = warnSpy.mock.calls.map(c => String(c[0] ?? ''));
-    expect(
-      messages.some(m => m.includes('"google"') && m.includes('without max_batch_tokens')),
-      'google should warn (it has fixed-cap models)',
-    ).toBe(true);
+    // Pre-v0.34.5 this test asserted that google STILL warned (waiting for
+    // someone to cap it). google now declares max_batch_tokens, so the
+    // contract flips: every native/openai-compat recipe is capped, and the
+    // missing-cap warning is silent for the entire first-party set.
+    const missingCapWarnings = messages.filter(m =>
+      m.includes('declares an embedding touchpoint') &&
+      m.includes('without max_batch_tokens'),
+    );
+    expect(missingCapWarnings).toHaveLength(0);
   });
 
   test('every recipe with empty models[] declares user_provided_models OR has openai-fast-path', () => {