feat(memory): implement stability stage

Co-authored-by: multica-agent <github@multica.ai>

apps/web/server/api/v1/coauthor/sessions/[id]/respond.post.ts

@@ -1,7 +1,7 @@
 import { z } from 'zod'
 import { buildJsonCompletion, parseJsonCompletion } from '~/server/utils/ai-provider'
 import { CoauthorQuestionSchema, StoredCoauthorQuestionSchema } from '~/server/utils/coauthor-question'
-import { requireStageTwoSafeText, requireStageTwoUser, stageTwoApiError, stageTwoDatabase, stageTwoRpc, stageTwoUuid } from '~/server/utils/stage-two-supabase'
+import { requireStageTwoSafeText, requireStageTwoUser, stageSixConsumeAiQuota, stageSixRecordAiUsage, stageTwoApiError, stageTwoDatabase, stageTwoRpc, stageTwoUuid } from '~/server/utils/stage-two-supabase'
 
 const MessageSchema = z.object({ role: z.enum(['user', 'assistant']), content: z.string().min(1).max(5000) })
 
@@ -34,8 +34,10 @@ export default defineEventHandler(async (event) => {
         schemaName: 'coauthor_question',
         jsonSchema: CoauthorQuestionSchema.toJSONSchema(),
       })
+      await stageSixConsumeAiQuota(user.id, null, 'coauthor-question')
       const controller = new AbortController()
       const timeout = setTimeout(() => controller.abort(), 30_000)
+      const startedAt = Date.now()
       let response: Response
       try {
         response = await fetch(completion.endpoint, {
@@ -45,8 +47,12 @@ export default defineEventHandler(async (event) => {
         clearTimeout(timeout)
       }
       if (!response.ok) throw createError({ statusCode: 502, statusMessage: 'The coauthor is temporarily unavailable.' })
+      const payload = await response.json()
+      await stageSixRecordAiUsage({
+        userId: user.id, campaignId: null, requestKind: 'coauthor-question', completion, payload, latencyMs: Date.now() - startedAt,
+      }).catch(error => console.error('[web] could not record coauthor question usage:', error))
       try {
-        const parsed = CoauthorQuestionSchema.safeParse(parseJsonCompletion(await response.json()))
+        const parsed = CoauthorQuestionSchema.safeParse(parseJsonCompletion(payload))
         if (parsed.success) question = parsed.data
       } catch {
         // Retry once when the provider ignores structured-output requirements.