fix(instrumentation-py): repair Chroma, Hugging Face, and Instructor OTel 2.x spans

[Nightingalelyy]

Summary

• restore Chroma workflow serialization through bounded SDK value normalization and end-to-end content privacy
• preserve complete Hugging Face batch content and validate a real local Transformers pipeline
• capture Instructor raw response schemas, calls, exact non-stream usage, and sync/async iterable results without SDK-object leakage
• add the required patch release intent for the four changed release-managed Python packages

Validation

• Chroma: 3 focused tests passed

• Hugging Face: 10 focused tests passed

• Instructor: 17 focused tests passed; 1 opt-in live test skipped

• affected respan-tracing tests: 60 passed

• affected respan-sdk tests: 7 passed

• four changed package wheels built successfully

• release inventory, release intents, formatting/import checks, and git diff checks passed

• all 14 linked examples completed under marker otel2-fix-py-group-15-20260817T084907Z

• Respan MCP: 14 connected traces and all 52 full records inspected; no errors, orphan parents, extra roots, or duplicates

• Full source and paired-example diffs were audited against every applicable document under contribution/; canonical span fields, OTel attribute shapes, release-intent/version ownership, and package boundaries have no remaining package-owned violation.

External follow-ups

• Instructor emits canonical llm.request.functions and gen_ai.completion.0.tool_calls. The platform retains both in semantic content and sets has_tool_calls=true, but its separate top-level tool arrays and model projection remain incomplete. The instrumentation intentionally does not add contribution-contract-forbidden aliases.

Companion examples

• fix(examples-py): validate Chroma, Hugging Face, and Instructor OTel 2.x spans respan-example-projects#63

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.