fix(snapshots): Add instrumentation logging to snapshot download stream

[NicoHinderling]

Add per-batch and lifecycle logging to the streaming ZIP download endpoint
to diagnose why 40K-image snapshot downloads fail with HTTP/2 stream errors
after ~20-25 seconds.

The streaming generator currently has zero observability — the Django span
closes in ~1s (before the generator runs), and when the stream dies
externally, nothing is logged. We need to see:

• How many batches/images are processed before the stream dies
• Whether there are long gaps between yielded chunks (idle timeout?)
• Fetch latency distribution from objectstore (slow reads?)
• Memory usage over time (OOM kill?)
• How the generator exits: completion, GeneratorExit (client/server disconnect), or unhandled exception

Log points added:

• stream_start — confirms generator is running, PID, memory baseline
• batch_complete — per batch: duration, cumulative progress, max yield gap, fetch latency p50/p99/max, RSS
• stream_complete — successful finish with totals
• stream_error — unhandled exception with traceback
• stream_finally — always fires with exit reason and final stats

This is temporary instrumentation to be removed once the root cause is found.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

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

^{Reviewed by Cursor Bugbot for commit b79a120. Configure here.}