Design an ELT pipeline to ingest JSON logs from millions of IoT devices. The data schema is evolving and prone to corruption. Describe the architecture from the ingestion layer (e.g., Kafka or Kinesis) to storage (e.g., Snowflake or S3). specifically addressing how you handle schema drift and data quality checks without blocking the pipeline. Explain the strategy for partitioning the data to optimize query performance for time-series analysis and how you would implement late-arriving data handling.