chore: add cached tokens props into RabbitMQ agent message

[forestileao]

Summary

• Add cache_read_tokens and cache_write_tokens fields to the AgentRunFinishedMessage
  protobuf message (fields 7 and 8)
• Propagate cached token counts through the AgentUsagePublisher protocol, UsagePublisher,
  and NoopUsagePublisher
• Pass cache_read_tokens and cache_write_tokens from RunUsage to the RabbitMQ message
  in _record_usage

Why

Previously, total_tokens in pydantic_ai's RunUsage only includes input_tokens + output_tokens — cached tokens were tracked in the DB (update_run_usage) but never
published to the RabbitMQ usage message. This means the SaaS usage service had no visibility
into prompt caching costs.

[superplanehq-integration]

👋 Commands for maintainers:

• /sp start - Start an ephemeral machine (takes ~30s)
• /sp stop - Stop a running machine (auto-executed on pr close)