Add Claude Agent SDK query instrumentation and tests

[eternalcuriouslearner]

Motivation

• Add instrumentation for claude_agent_sdk.query to emit invoke_agent spans and optionally capture message content, as noted in the changelog entry.
• Integrate with existing TelemetryHandler so the instrumentor delegates tracing/logging/metrics handling consistently with other GenAI instrumentations.

Description

• Implement patch.py that wraps claude_agent_sdk.query, injects SDK hooks, starts/stops an _QuerySession using TelemetryHandler.invoke_local_agent, and returns a _QueryStream wrapper to manage async iteration and error handling.
• Update the instrumentor __init__.py to construct a TelemetryHandler and call patch/unpatch instead of manually creating tracer/meter/logger instances.
• Add unit tests in tests/test_instrumentor.py to verify hooks are injected, user hooks are preserved, stream errors are recorded, and instrument/uninstrument behavior works as expected.
• Add a conformance scenario tests/conformance/query.py and a VCR cassette tests/cassettes/test_query_real_agent_span.yaml, and update tox.ini to include the conformance env and test commands for the Claude Agent SDK instrumentation.

Testing

• Ran unit tests for the instrumentation, including instrumentation/opentelemetry-instrumentation-genai-claude-agent-sdk/tests/test_instrumentor.py, and they passed.
• Executed the conformance scenario QueryScenario via tests/test_conformance.py with the provided VCR cassette using pytest, and it passed.
• Verified that the new patching/unpatching behavior does not break existing instrumentor lifecycle tests, and those tests passed.

---

Codex Task