feat: add complete MLflow UC tracing to all agent templates

[adamgurary]

Summary

Depends on: databricks/appkit#535

This instruments every executable agent surface discovered in app-templates with MLflow tracing and Unity Catalog setup, then adds behavior-derived conformance tests so future agent templates cannot be added without equivalent coverage.

• Traces complete agent execution trees across LangChain/LangGraph, OpenAI Agents SDK (including multi-agent handoffs), non-conversational and migration agents, support-agent pipelines, RAG/chat applications, MCP servers, and AppKit-generated agent templates.
• Captures bounded step inputs/outputs, errors, tool and retrieval calls, sub-agent handoffs, model metadata, token/cache usage, estimated cost when available, latency, time to first token, and streaming duration.
• Propagates W3C context and stable MLflow trace identity across framework, process, API, and AppKit boundaries.
• Adds UC experiment/table resources, authenticated OTLP configuration, quickstart/preflight setup, and template documentation.
• Adds executable-template discovery plus a shared trace contract/normalizer. The conformance suite derives its population from actual agent behavior, preventing a manually maintained allowlist from hiding omissions.

Verification

• LangChain TypeScript focused suites: 61 passed, plus tsc
• Multi-agent tracing suite: 9 passed
• Targeted Ruff correctness and format checks: passed
• Contract/discovery/sync suites: 55 passed
• Behavior-derived conformance: 24 passed in 212.62s
• All 39 contribution commits include DCO sign-off
• Pre/post-rebase tree hashes are identical (f8b0ee1a3050dbabeee540a30cb82bb6f2e8c020)
• OSS scrub: no internal URLs, workspace identifiers, warehouse identifiers, experiment identifiers, or credentials; remaining IDs are synthetic fixtures

Broad repository-wide Ruff output includes pre-existing template findings and was not used as the contribution gate; the targeted correctness and format checks for changed Python paths passed.

Live-environment limitations

• Managed Supervisor still needs live validation that it continues supplied W3C context or returns a persistently linkable remote trace identity.
• No live Databricks credential/profile was available for this verification run.
• The Apps Unity Catalog experiment resource used during earlier investigation returned HTTP 403.
• Therefore, this PR does not claim live Unity Catalog persistence was validated; framework instrumentation and behavior-derived conformance are covered locally.

[adamgurary]

@MarioCadenas Could you help route/review this companion to databricks/appkit#535? This repository has no CODEOWNERS routing for the touched paths, and fork contributors cannot request reviewers directly.