feat(audience): HttpTransport — background POST with gzip + retry (SDK-141)

[ImmutableJeffrey]

Summary

Adds HttpTransport and the Gzip helper that compresses event batches before upload.

What it does

• Reads batches of queued events from DiskStore, compresses them, and uploads them to the audience backend (/v1/audience/messages) on a background thread.
• Routes test API keys to sandbox and production keys to production.

Retry behaviour

• Backend accepts (2xx) → delete the batch.
• Backend rejects (4xx) → delete the batch (re-sending bad data won't help).
• Backend has problems (5xx) or network is down → keep the batch on disk and back off before retrying. Wait grows 5s → 10s → 20s → 40s, capped at 60s. Polling more often doesn't compound the wait — escalation only happens once the previous wait has passed.
• Local file read fails (e.g. permissions stripped) → drop the batch and report through the onError callback as FlushFailed, rather than silently crashing the background flush loop.

Caller-facing surface

• NextAttemptAt and IsInBackoffWindow expose retry timing so a scheduler outside the transport can decide when to call again.
• Failures fire an optional onError callback with one of three codes: ValidationRejected, FlushFailed, NetworkError. Exceptions thrown inside the callback are caught.

Linear: SDK-141

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f66dff6. Configure here.}