fix(examples-py): validate OpenAI, OpenAI Agents, and OpenInference OTel 2.x spans

[Nightingalelyy]

Summary

• make all 18 OpenAI SDK examples deterministic by default with optional live mode, exact run markers, bounded semantic roots, timeout aggregation, and explicit shutdown
• repair the OpenAI Agents example bridge, dependencies, direct-run imports, lifecycle, hosted-tool gates, deterministic contract scenarios, and aggregate runner
• add a five-scenario generic OpenInference example set covering chat/provider, tools, embedding, expected failure, streaming, and privacy
• link these examples to the local Group 20 instrumentation packages during validation

Validation

• example contract tests: OpenAI 6/6, OpenAI Agents 8/8, OpenInference 5/5
• all three example sets pass Ruff/format checks
• exact marker otel2-fix-py-group-20-20260817T182331Z: 47 traces and all 269 displayed record details inspected across this PR and its companion source PR
• OpenAI 18/18 and OpenInference 5/5 final processes passed
• the complete Agents runner previously passed under otel2-fix-py-group-20-20260817T174131Z with 23 traces / 402 inspected records; the replacement marker proves metadata visibility on all 23 Agents traces, while current live calls are externally blocked after authentication by missing provider key/managed credits

Known external follow-ups

• backend status/tool/stream/provider projections remain recorded separately
• shared batch-result task typing, marker indexing, and synthetic-root behavior belong to Respan core, not these examples
• hosted Responses-only Agent tools remain explicitly credential/resource-gated

Companion instrumentation PR: respanai/respan#386