feat(cli): stream model uploads with progress

[dylanuys]

Summary

• hash discriminator model archives incrementally instead of reading the entire file into memory
• stream model archives from disk during the presigned R2 upload
• show upload percentage, transferred size, throughput, and ETA while the PUT is active
• preserve the existing API flow, explicit content length, and five-minute timeout

Why

Large model uploads previously appeared idle during a single blocking PUT and temporarily consumed memory roughly equal to the archive size. Streaming keeps memory usage bounded and makes slow or stalled network paths visible without relaxing the existing timeout.

Validation

• pytest tests/test_miner_upload_streaming.py -q — 2 passed
• verifies incremental hashing
• verifies the request consumes a file-backed reader, retains Content-Length, reconstructs the exact payload, and reports completion progress