Attribute latency across server, authentication, network, and Azure dependencies

## Outcome

Performance investigations can identify whether latency originates in Azure MCP core, transport, authentication, networking, or downstream Azure services.

## In scope

- Define correlated timing spans for request receipt, authentication, routing, validation, Azure SDK calls, serialization, and response completion
- Separate server overhead from downstream and network time
- Preserve privacy-safe diagnostic dimensions
- Add benchmark and load-test correlation identifiers
- Provide investigation guidance for material regressions

## Out of scope

- Logging sensitive payloads or credentials
- Building general observability product workflows
- Owning downstream Azure service performance

## Acceptance criteria and budgets

- [ ] Representative benchmark and load-test requests expose correlated layer timings
- [ ] Unattributed time is less than 10% of end-to-end duration in controlled scenarios
- [ ] Timing instrumentation overhead remains below 5% of approved server-overhead baseline
- [ ] Diagnostics identify the dominant layer for an injected regression
- [ ] Sensitive-data handling follows the Safe Logging and Telemetry Epic

## Dependencies and risks

- Safe logging and telemetry boundaries
- Remote-host operational telemetry
- Performance workload and benchmark definitions
- Performance Epic #2412

## Evidence for closure

- Correlated timing model
- Regression investigation runbook
- Demonstrated attribution scenarios

Start with Performance Epic #2412, the Safe Logging and Telemetry Epic, remote-host operational telemetry, and the performance workload and benchmark definitions. Establish the correlated timing model and investigation runbook, then verify the stated attribution, overhead, regression-detection, and sensitive-data acceptance criteria in controlled scenarios.

Written by the indexing model from the issue text.