[FEAT] Better ScanTask sizing estimations

[jaychia]

Add capabilities for performing better sizing of our ScanTasks.

When reading data, we often just have a "size on disk" to try and understand how much data would actually be materialized in memory. This PR adds a new DataSizeEstimator trait that can be implemented to provide a mechanism for Daft to better estimate the size in memory of a ScanTask.

[codspeed-hq]

CodSpeed Performance Report

Merging #3257 will degrade performances by 35.89%

_{Comparing jay/better-scan-task-estimations (00dadad) with main (f290f40)}

Summary

❌ 1 regressions
✅ 16 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	jay/better-scan-task-estimations	Change
❌	test_iter_rows_first_row[100 Small Files]	177.3 ms	276.5 ms	-35.89%

[jaychia]

Closing in favor of #3302