feat: add Vapi memory cookbook example

[benfrank241]

Summary

Adds `applications/vapi-memory/` — a runnable FastAPI webhook server that gives Vapi voice assistants persistent caller memory via hindsight-vapi.

• `webhook_server.py` — FastAPI `/webhook` endpoint wired to `HindsightVapiWebhook`. Derives bank ID from `message.call.customer.number` so each caller gets their own memory.
• `simulate_call.py` — POSTs Vapi's two server event shapes (`assistant-request`, `end-of-call-report`) to the local webhook, so the full retain → recall cycle can be exercised end-to-end without a real Vapi account or phone number.
• `README.md` — walks through running the server, simulating calls, and wiring to a real Vapi assistant via ngrok.

Why draft

Depends on vectorize-io/hindsight#923 (the `hindsight-vapi` integration PR) merging and the package being published to PyPI. Once `pip install hindsight-vapi` works, this PR can be marked ready.

Test plan

• `python3 -m py_compile webhook_server.py simulate_call.py` — clean
• After hindsight-vapi releases: `pip install -r requirements.txt` succeeds
• End-to-end: `docker run ghcr.io/vectorize-io/hindsight:latest`, `python webhook_server.py`, then run `simulate_call.py end-of-call-report` once and `assistant-request` afterwards — confirm `assistantOverrides.model.messages[0].content` contains the recalled memory

🤖 Generated with Claude Code