fix(llma): trace summarization missing $ai_evaluation event type

Bug

The trace summarization pipeline (posthog/temporal/llm_analytics/trace_summarization/) is missing $ai_evaluation from its AI event type lists:

1. constants.py — AI_EVENT_TYPES tuple is missing $ai_evaluation
2. sampling.py — Two hardcoded WHERE event IN (...) SQL clauses are missing $ai_evaluation

This means any $ai_evaluation events within a trace are silently excluded from:

• Trace fetching (via queries.py which uses the AI_EVENT_TYPES constant)
• Trace sampling (via the hardcoded SQL in sampling.py)

The canonical Node.js list in nodejs/src/ingestion/ai/process-ai-event.ts includes $ai_evaluation, and the event type is used across the codebase (schema, taxonomy, usage reports, etc.).

Impact

Traces that contain evaluation events will have incomplete summaries — the evaluation results are dropped before the LLM summarizer sees them.

Expected behavior

$ai_evaluation should be included alongside the other AI event types in both the constants tuple and the sampling SQL queries.

Start in posthog/temporal/llm_analytics/trace_summarization/constants.py and sampling.py, then inspect queries.py to follow how the event lists are used. Add $ai_evaluation to the AI event tuple and both sampling SQL clauses, and verify that evaluation events are included in trace fetching and sampling.