fix: reduce retry attempts and use exponential backoff for presign/stage upload

[youngsofun]

Summary

Fixes retry behavior for presign upload/download and stage upload.

Changes

• MaxRetryAttempts: 20 → 5 for both PresignClient and stage upload
• Exponential backoff: replace linear attempts*100ms with min(1s * 2^n, 30s) (max ~31s total wait across 5 retries)
• Remove sinceStart timeout: the 900s wall-clock check incorrectly included actual file transfer time — a large file that took 800s to transfer would exhaust retries immediately after one failure
• Reduce timeouts: connectTimeout 600s→30s, write/readTimeout 900s/600s→300s (300s is still generous for large files; 600s connect timeout was meaningless)
• Fix false retry on NonRetryableHttpStatusException: short-circuit isRetryableIOException before checking error keywords, preventing response bodies containing "timeout" from triggering retries
• Remove unused RetryableHttpFailure class

Testing

Updated testDownloadStreamPresignedRetryExhaustionRaisesSQLException to reflect new attempt count (6) and extended timeout to 60s to accommodate exponential backoff.