Implement getKeyAsync with 100 ms batching and a single GET per batch

You are working in a single-threaded environment with concurrent I O (think event loop). Implement a function:

• getKeyAsync(key, callback) -> void

Within any 100 ms window, the implementation must batch multiple incoming key requests and issue exactly one HTTP request per batch:

• HTTP GET to path: /read?keys=k1,k2,...

You are given two helpers:

1. httpGetAsync(urlPath, callback) -> void, which asynchronously returns either an error or a JSON payload
2. parseKvs(json) -> Map<String, Integer>, which parses the JSON into a map from key to integer

After the batch response arrives, parse it and invoke each original per-key callback with its value.

Consistency requirement:

• A batch response may only serve callbacks for requests that arrived before the batch HTTP call was initiated. It must not serve later requests.

Also specify and design for the following:

• Handling duplicate keys within the same window
• Missing keys in the response
• Callback ordering
• Late arrivals that straddle batch boundaries
• Timeouts and retries
• Backpressure when requests outpace the network
• Error propagation to callbacks
• Graceful shutdown or drain behavior

Include a brief rationale about when you would choose callbacks versus multi-threading for this problem, and the trade-offs in complexity, latency, and throughput.