Avoid adding non-extra package with extra dependencies

[charliermarsh]

Summary

Previously, when we encountered foo[bar], we'd add a dependency on PubGrubPackage::Package for foo, and then PubGrubPackage::Extra for foo[bar].

Later, when we ask for the dependencies of the PubGrubPackage::Extra, we add PubGrubPackage::Package for foo, and PubGrubPackage::Package for foo[bar]. This is an intentional strategy because it ensures that PubGrub "knows" that these have to be solved to the same version as early as possible.

It turns out that the first part here ("add a dependency on PubGrubPackage::Package for foo") is suboptimal, because it means PubGrub might try to solve just foo without realizing that it also has to accommodate all the constraints from the extra.

Instead, we now add just PubGrubPackage::Extra for foo[bar], and defer adding the base package. It looks like this leads to a far more efficient solve for Airflow.

[codspeed-hq]

CodSpeed Performance Report

Merging #9540 will improve performances by 77.1%

_{Comparing charlie/transitive (042f88a) with main (8126a5e)}

Summary

⚡ 1 improvements
✅ 13 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	charlie/transitive	Change
⚡	resolve_warm_airflow	2 s	1.1 s	+77.1%

[charliermarsh]

The gain actually seems real? At least to some degree:

❯ hyperfine "./target/release/uv pip compile req.in" "./target/release/transitive pip compile req.in" --warmup 10 --runs 50
Benchmark 1: ./target/release/uv pip compile req.in
  Time (mean ± σ):     198.6 ms ±   6.0 ms    [User: 224.9 ms, System: 244.6 ms]
  Range (min … max):   189.8 ms … 213.9 ms    50 runs

Benchmark 2: ./target/release/transitive pip compile req.in
  Time (mean ± σ):     149.4 ms ±   4.1 ms    [User: 160.6 ms, System: 182.2 ms]
  Range (min … max):   144.9 ms … 173.2 ms    50 runs

Summary
  ./target/release/transitive pip compile req.in ran
    1.33 ± 0.05 times faster than ./target/release/uv pip compile req.in

[charliermarsh]

This change is sound but I clearly need to modify some of how we interpret the downstream graph in pip compile.

[charliermarsh]

I reasoned through these and most of them can be removed.

[charliermarsh]

If we just use name, we can rely on warn_user_once.

[konstin]

wow wouldn't have thought about that trick for such a speed!