[BUG] Avoid reconstructing sql query in read_sql

[colin-ho]

When Daft executes a read_sql scan task, it calls the def read_sql function in table_io.py. This function then calls the .read method on the SQLConnection object. However, the .read method will reconstruct the sql query and add another layer of subqueries, which is unnecessary.

This is because the .read method constructs a sql query given additional predicates, projections, and limits, then executes it. However, the scan task is already given a constructed query with pushdowns applied, so this reconstruction is unnecessary.

This PR removes the .read method and instead exposes the execute_sql_query method. Having .read do construction and execution together is confusing.

[codspeed-hq]

CodSpeed Performance Report

Merging #2818 will degrade performances by 83.68%

_{Comparing colin/unncessary-sql-subquery (ae62105) with main (d30e62a)}

Summary

❌ 1 regressions
✅ 15 untouched benchmarks

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

Benchmarks breakdown

	Benchmark	main	colin/unncessary-sql-subquery	Change
❌	test_show[100 Small Files]	48.7 ms	298.6 ms	-83.68%