Parallelize chunk uploads in record tests for better throughput

[Konboi]

Why

When uploading millions of test records (e.g. with the raw profile), the record tests command is too slow because event chunks are posted sequentially. With the default chunk size of 1000, uploading 1M records requires ~1000 HTTP round-trips, each blocking until the response arrives. The main thread idles during network I/O instead of parsing the next chunk.

What

Use concurrent.futures.ThreadPoolExecutor (3 workers) to POST event chunks concurrently instead of sequentially:

• The first chunk is sent synchronously to handle no-build mode, where the response sets the session ID for subsequent requests
• All subsequent chunks are submitted to the thread pool and uploaded in parallel
• Parsing continues on the main thread while uploads happen in the background
• After all chunks are submitted, we wait for all futures to complete and propagate any exceptions

This overlaps network I/O with parsing and reduces wall-clock time by approximately 3x for large uploads.

Please Review Here

• Thread safety: payload() (which mutates count) runs only on the main thread. send() runs in worker threads but only reads self.session (set after the first sync call) and sets is_observation (same value every response). requests.Session is thread-safe for concurrent requests.
• The first-chunk-sync pattern ensures no-build mode works correctly since the response establishes the session ID used by all subsequent requests.