Fix AI narration truncation: reduce chunk size to 8, increase maxTokens to 8192

Chunk 1 returned 9 of 10 expected items — the response was truncated at
4096 tokens. Reducing chunk size from 10 to 8 and doubling maxTokens to
8192 gives adequate headroom for the conversational prompt style which
produces longer narratives per discovery.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: schwaaamp

supabase/functions/ai-engine/engines/pattern-spotter.ts

@@ -1037,7 +1037,7 @@ export async function spotPatterns(
   const allToNarrate = [...discoveriesToSurface, ...observationsToSurface];
   const newPatterns: Array<Record<string, unknown>> = [];
   let aiCallSucceeded = false;
-  const NARRATION_CHUNK_SIZE = 10; // Max discoveries per AI call to avoid truncation
+  const NARRATION_CHUNK_SIZE = 8; // Max discoveries per AI call to avoid response truncation
 
   if (allToNarrate.length > 0) {
     const systemPrompt = buildPatternDetectionSystemPrompt();
@@ -1069,7 +1069,7 @@ export async function spotPatterns(
           systemPrompt,
           userPrompt,
           temperature: 0.3,
-          maxTokens: 4096,
+          maxTokens: 8192,
           responseFormat: 'json',
         });