Support retries for hard node.js network failures

[thecodedrift]

Making fetch requests from within node has historically had "glitches" such as the underlying Socket failing with ECONNRESET. Because mongo-data-api talks to a realm app over HTTPS, it's particularly prone to this problem.

1. Select a low-overhead retry library (possible list https://npmtrends.com/p-retry-vs-promise-retry-vs-retry)
2. Wrap the fetch calls in retry to minimize connection errors
   1. Is this configuration done at the top level (new MongoClient) or at the individual method level (.findOne(filter, options))
   2. Should backoff be configurable, or should we focus on defaults that make sense for a Realm App
3. Create suitable mocks to simulate the socket failing

[thecodedrift]

This sounded like a good idea on the surface, but we don't know how the api is being used. For example, an aggregation pipeline may not have retries, but then an update may need several. Configuring backoffs etc is probably beyond the scope of this module when there's great libraries like retry available.