Release v0.1.15: SSE Performance — Extract retryCount in SQL

What changed

The SSE stream endpoint for real-time pipeline monitoring was selecting the entire inputData JSONB column on every poll tick (every 3 seconds per connection), then immediately discarding it after extracting retryCount. Since inputData stores full AI conversation history, this meant megabytes of data transferred and deserialized per tick.

We now extract retryCount directly in SQL using a JSONB expression, so only the integer is transferred.

Impact

• 50–90% reduction in database data transfer per active pipeline connection
• Significant CPU savings from eliminating JSONB deserialization on every poll
• No API changes — retryCount output remains identical
• Security improvement: sensitive credentials in inputData never touch the application server during streaming

Technical

Replaced SELECT agentJobs.inputData + JS map with a SQL expression: sql<number | null>\(${agentJobs.inputData}->>'retryCount')::int``. The result shape is correct at query time, eliminating the intermediate transformation step.