Queue limit errors downgraded to warnings

Queue size limit errors were being logged at error level across the system, but these are expected validation rejections — not bugs. When a task can't be triggered because the queue is full, that's working as intended, not a failure.

The fix adds a logLevel property to ServiceValidationError in both the webapp and run-engine packages. All queue limit throws now explicitly set logLevel to "warn". The schedule engine detects these queue limit failures and categorizes them separately from system errors. The redis-worker reads the logLevel from thrown errors and logs at the appropriate level.

In the webapp's schedule engine, failed triggers now return an errorType field distinguishing "QUEUE_LIMIT" from "SYSTEM_ERROR". The schedule engine then logs queue limit warnings separately from other task failures, and reports them with a distinct error_type in metrics. Some noisy info-level logs in the run-queue were also downgraded to debug.

This reduces alert fatigue and makes logs more meaningful — errors represent actual problems, while warnings represent expected rejections.