doc: add KQL query library section to telemetry.md #89

Objective

Add a ready-to-use KQL query section to docs/telemetry.md so users who send eval traces to Azure Monitor / Application Insights can immediately query their data.

Background

When eval traces land in App Insights (via OTLP ingestion — see #88), all span attributes are available as custom dimensions in the dependencies and requests tables. Users need practical KQL examples to extract value from this data.

Proposed Queries

Add a new section between 'Sending Traces to Azure Monitor' and 'Evaluation Tracing vs. Agent Execution Tracing' with queries like:

1. Slowest evaluation rows — top N eval_item spans by duration
2. Failed evaluators — filter by agentops.eval.evaluator.passed == false
3. Pass rate over time — trend of agentops.eval.pass_rate from root spans
4. Token usage per run — sum of gen_ai.usage.input_tokens + output_tokens
5. Evaluator score distribution — histogram of agentops.eval.evaluator.score grouped by evaluator name

Format

Each query should include:

• A short description of what it answers
• The KQL query itself (ready to paste into App Insights → Logs)
• Expected table/column names based on OTLP → App Insights mapping

Acceptance Criteria

• KQL section added to docs/telemetry.md
• At least 5 practical queries covering performance, failures, quality trends, and token usage
• Queries reference the correct App Insights tables (dependencies, requests, customDimensions)
• Tested against real App Insights data if test: validate eval traces in Azure Monitor (App Insights + GenAI blade) #88 is completed first

Dependencies

• Ideally done after test: validate eval traces in Azure Monitor (App Insights + GenAI blade) #88 validates that traces actually arrive in App Insights correctly
• Can be written speculatively based on known OTLP → App Insights mapping