feat: add telemetry engine for distributed tracing

[TonsOfFun]

Summary

Adds a comprehensive telemetry system for collecting and reporting agent traces to hosted or self-hosted observability endpoints. This is competitive with Phoenix Arize but built natively for Rails.

Features:

• Trace collection with spans for generation lifecycle (prompt, llm, tool, error)
• Token usage tracking (input, output, thinking tokens)
• Tool call tracing with timing and results
• Error recording with backtraces
• Async HTTP reporting with batching and background flush
• Configurable sampling rate for production traffic
• Auto-instrumentation via railtie when enabled

Configuration via YAML:

telemetry:
  enabled: true
  endpoint: https://api.activeagents.ai/v1/traces
  api_key: your-api-key
  sample_rate: 1.0
  batch_size: 100
  flush_interval: 5

Or programmatically:

ActiveAgent::Telemetry.configure do |config|
  config.enabled = true
  config.endpoint = "https://api.activeagents.ai/v1/traces"
  config.api_key = Rails.application.credentials.dig(:activeagents, :api_key)
end

Files added:

• lib/active_agent/telemetry.rb - Main module with NullSpan for disabled state
• lib/active_agent/telemetry/configuration.rb - Configuration options
• lib/active_agent/telemetry/span.rb - Span with attributes, tokens, events
• lib/active_agent/telemetry/tracer.rb - Trace management with thread-local context
• lib/active_agent/telemetry/reporter.rb - Async HTTP reporter with batching
• lib/active_agent/telemetry/instrumentation.rb - Auto-instrumentation module
• docs/framework/telemetry.md - Comprehensive documentation

Test plan

• Test telemetry disabled by default (no impact on existing behavior)
• Test YAML configuration loading
• Test programmatic configuration
• Test span creation and hierarchy
• Test token tracking
• Test error recording
• Test async reporting doesn't block generation
• Test sampling rate works correctly

🤖 Generated with Claude Code