feat(host): Add retry limits and exponential backoff to preimage fetching

[RealiCZ]

Problem

The get_preimage method in OnlineHostBackend had an infinite retry loop with no timeout or backoff mechanism. This caused several issues:

1. Infinite retries: If the Beacon node is unreachable or blob data is unavailable, the loop would retry forever
2. Resource exhaustion: Continuous failed attempts consume CPU and network resources
3. No backoff: Immediate retries create unnecessary load on external services
4. Busy waiting: When no hint is available, the loop would spin without delay

This became apparent when encountering the error:
Failed to fetch blobs: Preimage oracle error: Channel is closed.

Solution

Added comprehensive retry protection following the project's existing retry patterns from crates/supervisor/service/src/actors/utils.rs:

Key Changes

1. Timeout Protection: 30-second overall timeout per preimage fetch
2. Retry Limit: Maximum 100 retry attempts before failing
3. Exponential Backoff: Millisecond-level delays suitable for high-frequency operations
   • Progression: 100ms → 200ms → 400ms → 800ms → 1.6s → 3.2s (capped at 5s)
   • Uses saturating_pow and saturating_mul to prevent overflow
4. Improved Logging:
   • Error logs show attempt count and total retries
   • Warn logs display the actual backoff delay
5. Graceful Degradation: Handles missing hints with appropriate delays

Implementation Details

• Created preimage_backoff_delay() function mirroring the project's backoff_delay() pattern
• Millisecond-level backoff (vs. second-level) appropriate for preimage operations
• Added early return for already-cached preimages
• Structured logging with ?delay format for better observability

Testing

• Code compiles without errors
• Follows existing project patterns and style
• No linter errors

Behavior Changes

• Success case: No change - cached preimages return immediately
• Retry case: Now includes exponential backoff delays (max 100 retries)
• Timeout case: Returns PreimageOracleError::Timeout after 30 seconds
• Failure case: Returns error after 100 attempts with detailed logging

Related Issues

Addresses scenarios where Beacon node connectivity issues or missing blob data would cause the host to hang indefinitely.