fix(instrumentation-py): repair LiveKit, LlamaIndex, and Marqo OTel 2.x spans

[Nightingalelyy]

Summary

• mark LiveKit LLMStream calls as streaming, derive provider identity from the LiveKit plugin/model instead of the gateway host, preserve canonical tool schema/calls, and remove temporary transport attributes before export
• classify only actual LlamaIndex BaseTool.call/acall executions as tools; keep orchestration phases as tasks and cover a registered real ReAct runtime with exactly one multiply_numbers execution
• serialize Marqo SDK identities into stable bounded data, redact endpoint userinfo/query/fragment secrets, and cover current real-client success plus 503 failure without hiding duplicate spans
• add patch release intent for all three instrumentation packages

Validation

• focused tests: LiveKit 8/8, LlamaIndex 24/24 plus one credential-gated skip, Marqo 3/3 (35 passed, 1 skipped)

• Ruff check and format-check passed for every changed Python file

• wheels built successfully for all three packages

• release inventory, release-intent, missing-intent, and diff checks passed

• all validation environments resolved the three instrumentations and local Respan core packages through editable local direct_url.json entries

• corresponding examples passed 12/12 with marker otel2-fix-py-group-17-20260817T111735Z

• Respan MCP: 12 connected traces / 58 full records; one root per tree, matching counts, no duplicate IDs or orphan parents; every full record inspected for hierarchy, type, content, model/provider, usage, tools, errors, grouping, and exact marker

• final LiveKit correction rerun otel2-fix-py-group-17-livekit-clean-20260817T112900Z: 5 traces / 7 records, provider openai, stream=true, exact usage, complete tool schema/call, and no internal transport-key leak

• Contribution re-audit against every current document under contribution/ passed after 2b64d6c: LlamaIndex text-completion events use the permitted text log type and canonical llm.request.type=chat.

• Fresh correction validation: LiveKit 8/8, LlamaIndex 24/24 plus 1 skipped, and Marqo 3/3 passed; LlamaIndex wheel, release inventory/intents, formatting/import, and diff checks passed; marker otel2-pr-contract-corrections-20260818T133300Z produced 1 connected LlamaIndex trace / 2 full records, both inspected.

External follow-ups

• full-detail convenience tools / tool_calls arrays remain empty despite canonical tool attributes/content
• LlamaIndex agent model and task/workflow display names require backend/UI derivation under the common-span contract
• Marqo native ERROR spans retain the 503 exception but ingest as success/200 with trace error_count=0

Companion examples

• fix(examples-py): validate LiveKit, LlamaIndex, and Marqo OTel 2.x spans respan-example-projects#65

[greptile-apps]

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