dev: separate Langfuse benchmarks and Codex setup

[kennethbruskiewicz]

Separates developer-facing Langfuse tooling from the runtime instrumentation merged in #55.

• Adds deterministic offline turn benchmarks and a separately invoked live acceptance probe.
• Preserves the earlier sanitized acceptance reports for review; no new live traces were sent for this change.
• Adds the optional Codex configuration helper and documents that it is independent of TCT runtime activation.
• Updates the offline benchmark to use the new telemetry module.

Validation: five benchmark/acceptance unit tests passed; spelling and whitespace checks passed. The live acceptance probe and Codex configuration helper were not executed for this revision.

Draft for separate review of developer-tooling scope.