feat(tracing): add advanced sampling strategies

[CalvinAllen]

Summary

Add more sophisticated sampling strategies beyond the current simple ratio-based sampling. This enables cost-effective telemetry while ensuring important data is captured.

Proposed Sampling Strategies

1. Error-Biased Sampling

Always capture errors at 100%, sample successes at a lower rate.

config.Sampling = new ErrorBiasedSampler(
    errorSampleRate: 1.0,      // 100% of errors
    successSampleRate: 0.1     // 10% of successes
);

2. Adaptive Sampling

Automatically adjust rate based on volume to stay within a target throughput.

config.Sampling = new AdaptiveSampler(
    targetSpansPerMinute: 1000,
    minSampleRate: 0.01,
    maxSampleRate: 1.0
);

3. Per-Operation Sampling

Different rates for different operation types.

config.Sampling = new PerOperationSampler(new Dictionary<string, double>
{
    ["CommandExecution"] = 1.0,    // Always capture commands
    ["DocumentChange"] = 0.01,     // Rare for noisy events
    ["*"] = 0.1                    // Default for others
});

4. Parent-Based Sampling

Respect parent span's sampling decision for distributed traces.

config.Sampling = new ParentBasedSampler(
    rootSampler: new RatioSampler(0.1)
);

Implementation Notes

• Implement Sampler base class with ShouldSample() method
• Integrate with OpenTelemetry's sampler pipeline
• Add sampling decision to span attributes for debugging