This commit introduces an in-memory caching mechanism for the Montpellier
real-time transit data CSV. Instead of fetching and parsing the CSV on
every API request, the data is now cached for 1 minute.

Key changes:
- Fetched CSV data is stored in a module-level cache (`cachedData`).
- A timestamp (`lastCacheTime`) tracks the age of the cache.
- If a request arrives and the cache is older than 1 minute or empty,
  the data is re-fetched and the cache is updated.
- An initial attempt to populate the cache is made when a function
  instance starts.
- The `api/query.js` handler now uses this caching logic.
- Added comments explaining the in-memory cache behavior on Vercel's
  serverless environment, noting it's per-instance and suggesting
  Vercel KV and Cron Jobs for persistent, shared caching.
- Introduced a test file `api/query.test.js` with unit tests for
  the caching logic (initial fetch, cache hit, cache expiration) and
  data filtering.
- Modified `api/query.js` to allow testability of cache internals
  when `NODE_ENV=test`.

This change aims to reduce redundant downloads and improve API response
times for frequent requests to the same function instance.