docs(truefoundry): refresh Python client tracing setup - #1069

Summary

Refresh the TrueFoundry Python client guide to use published traceai-openai 0.1.11, fi-instrumentation-otel 1.2.2 and OpenAI 3.42.0 instead of retaining limitations already fixed in released packages.

• Explain the requested model retained on streamed/failed calls and the available error-redaction controls. Preserve the distinction between the OpenAI-compatible provider label and the gateway's actual upstream provider.
• Put credentials, the authorized gateway model and privacy controls before the runnable example; flush spans even when a request fails.
• Share an explicit TraceConfig between the full example's client and agent/tool spans, so environment controls are read when the example is initialized.
• Keep this a client-only setup, link the upstream native-export configuration, and remove outdated verification-diary prose from the example README. No new integration package or gateway-wide compatibility claim.

Regression coverage

The old recipe fails the streamed-model, failed-model and error-redaction regressions. The updated suite covers real released OpenAI clients at 1.69.0 and 3.42.0, sync/async success and failure, streaming, the page snippet's real OTLP exporter, and the README's agent/tool hierarchy. Tests deny non-loopback connections and use synthetic credentials. The JSON wire fixture now uses OTLP-required numeric enum values.

A scoped CI workflow runs the client suite on both SDK versions. The exact requirements install, published source-distribution builds, docs build, integration checks, link audit and local docs rendering were also checked. No live gateway or Future AGI account is required by these tests.

Refs TH-8304. All installation pins are already published; no SDK release is required for this documentation update.

Generated with Claude Code