[deployer]: remove unbounded layered S3 retries

[GarmashAlex]

Why were these changes needed?

S3 uploads had two unbounded retry layers at the same time:

1. SDK retry config used with_max_attempts(u32::MAX),
2. upload_with_retry also had an infinite manual retry loop.

This made failures hard to diagnose and could cause very long or effectively unbounded wait times.

What does this change do?

• Caps S3 SDK retries to a finite value (max_attempts = 5).
• Removes the manual infinite retry loop from upload_with_retry.
• Keeps retry responsibility in one place (AWS SDK policy).
• Propagates upload errors to callers instead of retrying forever at the wrapper level.

[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.}

[cursor]

Wrong S3Operation variant used for PutObject error

Medium Severity

The upload_with_retry function performs a put_object call but tags the error with S3Operation::HeadObject. This causes any upload failure to be reported as a HeadObject error, making it very difficult to diagnose S3 upload issues. The S3Operation enum also lacks a PutObject variant, so one needs to be added.