perf: flat SQL for single-op pipelines and group_by+summarise

[hugobarauna]

Summary

Skip CTE wrapping for common pipeline shapes:

• Single-op pipelines (e.g. filter or mutate alone) emit flat SQL instead of WITH __s0 AS (...) SELECT * FROM __s0
• group_by + summarise (the most common aggregation) emits a single SELECT ... GROUP BY without CTEs
• Table sources are referenced by name directly instead of wrapping in (SELECT * FROM "table") __src

Benchmark (1M rows, Apple M2 Pro, Benchee)

Operation	main	this PR	Change
compute: filter	0.98ms (1018 ips)	0.94ms (1066 ips)	+5%
compute: mutate	0.66ms (1507 ips)	0.59ms (1701 ips)	+13%
e2e: summarise→to_rows	7.79ms (128 ips)	6.11ms (164 ips)	+28%
e2e: filter→to_rows	361ms (2.77 ips)	349ms (2.87 ips)	+4%

Benchmark script: https://gist.github.com/hugobarauna/369f68acf4292f27309782ba774b8988

Example

Before:

WITH
  __s0 AS (SELECT * FROM (SELECT * FROM ("__dux_123") __src) __src WHERE quantity > 50)
SELECT * FROM __s0

After:

SELECT * FROM "__dux_123" WHERE quantity > 50

[cigrainger]

Hey @hugobarauna thank you for this. Looks like a pretty straightforward win! Crazy few days over here but I'll have a proper look Monday.

[cigrainger]

Thank you! <3