Added `onDiskSizeBytes` field to `CollectionStats` to track the compressed on-disk size of collections, separate from the existing `logicalDataSizeBytes` which represents uncompressed in-memory size.

• Modified join cost estimator to use on-disk size instead of logical size for I/O cost calculations in:
• Collection scan fragment costing
• Index scan fragment costing
• Index nested loop join fragment costing
• Updated all test fixtures and expected test outputs to include the new field
• Test helper now sets on-disk size to 1/4 of logical size to simulate compression

This change improves cost estimation accuracy by using actual compressed storage size when calculating disk I/O costs for query execution plans.