Add production-grade randomization to error recovery retry system

[milliondreams]

Description

The error recovery system currently uses a simplified random implementation for jitter calculations in retry delays. This should be replaced with a proper randomization implementation for production use.

Current Implementation

In crates/mandrel-mcp-th/src/error_handling/recovery.rs, there is a placeholder rand module that always returns 0.5:

mod rand {
    pub fn random<T>() -> T
    where
        T: From<f64>,
    {
        // Placeholder implementation
        T::from(0.5)
    }
}

Required Changes

1. Add rand crate dependency to Cargo.toml
2. Replace the placeholder implementation with proper randomization
3. Ensure thread-safe random number generation for concurrent retry operations
4. Add tests to verify jitter behavior is properly randomized

Priority

Enhancement for production deployment

Labels

• enhancement
• testing
• code-quality
• performance

Related Code

• File: crates/mandrel-mcp-th/src/error_handling/recovery.rs
• Function: RetryExecutor::calculate_delay()
• Context: Jitter calculation for exponential backoff retry delays