Preserve completed runs from overwrite

Reduce validation GPU memory and reuse LMDB handles

Fix LMDB reuse during validation

Support pre-split parquet training inputs

Reduce parquet split-map memory

Sync DeepGenopix package snapshot

Use proportional family holdouts for parquet ETL

Update src/deepgenopix/notebook_support.py from ml-intern branch

Update src/deepgenopix/cli.py from ml-intern branch

Upload src/deepgenopix/etl_streaming.py

Upload src/deepgenopix/cli.py

Upload src/deepgenopix/__init__.py

Upload src/deepgenopix/quant.py

Upload src/deepgenopix/analysis.py

Upload src/deepgenopix/inference.py

Upload src/deepgenopix/trainer.py

Upload src/deepgenopix/model.py

Upload src/deepgenopix/dataset.py

Upload src/deepgenopix/etl.py

Upload src/deepgenopix/io_utils.py

Upload src/deepgenopix/pixelizer.py

Upload src/deepgenopix/config.py

Upload src/deepgenopix/__init__.py