fix+feat: morning QA fixes, worklist pagination, misc sidecar improvements

- caller-resolution: drop cache, use indexed phone filter (lead.contactPhone.primaryPhoneNumber.like)
- worklist: externalize page size (WORKLIST_PAGE_SIZE × WORKLIST_MAX_PAGES), paginate getMissedCalls/getAssignedLeads/getPendingFollowUps
- maint: unlock-agent, force-ready, backfill-caller-resolution, clear-analysis-cache, fix-timestamps
- ozonetel agent.service: force logout+re-login on "already logged in"
- ai chat: context expansion
- livekit-agent: updates
- widget: session handling
- masterdata: clinic list cache

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

src/config/configuration.ts

@@ -22,6 +22,16 @@ export default () => ({
     missedQueue: {
         pollIntervalMs: parseInt(process.env.MISSED_QUEUE_POLL_INTERVAL_MS ?? '30000', 10),
     },
+    worklist: {
+        // Per-page fetch size from the platform GraphQL endpoint. Tuned to
+        // balance response size vs. page count. Platform's Relay pagination
+        // typically caps at 100–200 per page.
+        pageSize: parseInt(process.env.WORKLIST_PAGE_SIZE ?? '50', 10),
+        // Hard ceiling on pages fetched per poll. Safety valve against
+        // unbounded cost when a tenant has thousands of pending callbacks.
+        // maxPages * pageSize = effective worklist size.
+        maxPages: parseInt(process.env.WORKLIST_MAX_PAGES ?? '10', 10),
+    },
     ai: {
         provider: process.env.AI_PROVIDER ?? 'openai',
         anthropicApiKey: process.env.ANTHROPIC_API_KEY ?? '',