Enhance ADK BigQuery Agent Analytics Plugin

[haiyuan-eng-google]

1. Event-Specific BigQuery Views (Update in BQ AA SDK)
   To simplify data consumption, the goal is to move away from a single generic table that requires complex parsing.

• Per-Event Views: Create individual views for each event type (e.g., LLM_REQUEST, LLM_RESPONSE, TOOL_CALL).
• Data Unnesting: These views should automatically unnest specific event-type attributes into dedicated columns.
• Standardized Headers: Every view must retain core identity fields: ID, agent_name, and timestamp.

2. Advanced Tree Structure & Error Visibility (Update in BQ AA SDK)
   The "Tree Structure" used in the Python SDK needs to be more robust regarding failure states.

• Explicit Error Reporting: Ensure that errors occurring during tool calls are explicitly highlighted at the correct node in the execution tree.
• Failure Context: Capture the specific reason a tool failed.

3. Specialized Tracing for Human-In-The-Loop (HITL) (Update in BQ AA Plugin)
   Standard latency metrics are insufficient for actions requiring human interaction.

• Distinguish Special Invocations: Differentiate technical tool calls from functions waiting for user approval or authentication.
• Specific Event Tracking: Track adk_request_confirmation (for human validation) and adk_request_credential (for OAuth2 flows).
• New Metrics: Introduce a "Human Interaction Count" and separate latency tracking for time spent waiting for human responses.

4. Origin & Tool Provenance Differentiation (Update in BQ AA Plugin)
   Logs should distinguish between different call origins to support complex agent architectures.

• Differentiate Entry Types: Identify whether a call originates from Local Tools, Local Sub-agents, or Remote Entities (via MCP or A2A).

5. Schema Management & Auto-Upgrades (Update in BQ AA Plugin)

• Auto-Upgrade: Implement a mechanism for the plugin to automatically add new columns to the BigQuery table as the schema evolves.
• Schema Governance: Introduce versioning and verification to ensure data integrity across different ADK versions.