feat: Add Restart Processing + fix Teams link-unfurling cancel/restart loop

[yaqi-lyu]

Summary

This branch adds a Restart Processing capability and fixes a critical bug where Teams' link-unfurling (URL prefetch) was causing an infinite cancel → restart → cancel loop.

---

Bug Fix: Teams Link Unfurling Loop

Root Cause

Teams automatically sends GET requests to URLs in notification cards for link-preview generation. Since CancelProcessing and RestartProcessing both accepted GET requests and executed their actions immediately, this caused:

"started" notification with cancelUrl → Teams prefetches cancelUrl (GET) → auto-cancel
"cancelled" notification with restartUrl → Teams prefetches restartUrl (GET) → auto-restart
→ infinite loop

This was only triggered after the restartUrl was added to "cancelled" notifications (the new feature in this branch), which closed the loop.

Fix

• GET → returns a confirmation page (no action taken, safe for Teams prefetch)
• POST → executes the actual cancel/restart

For RestartProcessing GET, checks in-memory cache first:

• If already running → shows "Already Running" page immediately (no confirmation step needed)
• If not running → shows confirmation page

---

New Feature: Restart Processing

Adds RestartProcessing.js — an HTTP endpoint that re-queues a transcript for processing after cancellation or failure.

Concurrency protection (3 layers):

1. In-memory cache check (GET, instant) — shows "Already Running" if same-instance cache has a live execution
2. Container Apps API live check (POST) — full API scan before queueing, catches cross-instance cases
3. Queue-level dedup (restartId) — prevents Azure Queue redelivery from firing twice

---

Changes

New files

• azure-function/src/functions/RestartProcessing.js — HTTP trigger for restart
• azure-function/src/htmlResponseCard.js — shared HTML response card + new renderConfirmationCard

Modified files

CancelProcessing.js

• GET → confirmation page (breaks Teams unfurl loop)
• POST → cancel action (existing logic, now with security hardening)
• Infrastructure IDs (subscriptionId, resourceGroup, jobName) read server-side from env vars, not from URL params
• Cancel success page: removed Restart button (users restart via Teams notification)
• Added buildRestartUrl helper to pass restart URL to notifications

ProcessTranscriptQueue.js

• Added findActiveExecutionForMeeting() — lets RestartProcessing check if a job is already running for a meeting
• Added restartTrigger / restartId dedup path — restart messages use per-click unique ID to prevent Azure Queue redelivery, while still allowing multiple legitimate restarts

entrypoint.sh

• Cancel checker background process polls every 15s; on cancellation signals parent process group
• on_sigterm checks CheckCancellation endpoint before sending cancelled notification (avoids mislabelling platform shutdowns as user cancels)
• exit 0 on confirmed user cancel so Container Apps Job status = Succeeded (not Failed), keeping Stopped/Failed/Succeeded states semantically distinct

sendNotification.js

• Passes restartUrl in "cancelled" and "failed" notifications
• Passes cancelUrl in "started" notifications