fix(examples-py): validate MCP, Microsoft Agent Framework, and Milvus OTel 2.x spans

[Nightingalelyy]

Summary

• make the MCP examples portable on the verified 1.x SDK line, preserve the exact audit marker, return bounded tool/resource/prompt results, add an expected connection-failure case, and shut down explicitly
• give Microsoft Agent Framework examples deterministic real-framework success/failure coverage plus an optional live gateway tool round with bounded lifecycle handling
• make Milvus lifecycle/data results JSON-native, avoid the successful-run timestamp RPC mismatch, add a deterministic missing-collection case, preserve marker identity, and shut down explicitly

Instrumentation companion: respanai/respan#384

Validation

• All 10 example processes pass with local editable packages and repository .env
• Shared marker: otel2-fix-py-group-18-20260817T121654Z
• MCP: 4/4 processes, 4 traces, 14 full records
• Microsoft Agent Framework: 3/3 processes, 3 traces, 16 full records, including live gateway success
• Milvus: 3/3 processes, 3 traces, 19 full records
• Every trace tree and all 49 full records were inspected for hierarchy, types, model/provider, usage, inputs/outputs, tools, errors, grouping, and duplicates
• Ruff check, format-check, Python compilation, and diff checks pass for all changed examples

Known external residuals

• the platform still projects native MCP and Milvus OTel ERROR spans as success/200 after ingestion
• Agent Framework common-only agent model/token display and canonical-tool top-level arrays remain platform projection work
• native Agent Framework bookkeeping spans do not expose semantic payload upstream; the examples provide bounded outer workflow/task content