Avoid repeated fragment complexity analysis

[ydah]

Improve query complexity analysis for recursively reused fragments.

QueryComplexity currently visits each fragment spread as an independent expanded tree. A fragment graph such as:

fragment F2 on Node {
  a { ...F1 }
  b { ...F1 }
}

revisits the same nested selections for every path, causing analysis time to grow exponentially with the fragment level.

This PR calculates concrete object selections as a DAG and memoizes shared fragment complexity. Built-in QueryComplexity and MaxQueryComplexity analyzers use the precomputed result, while abstract selections and custom analyzer subclasses retain the existing visitor path.

Multiplex analyzers decide visitation per query, preventing an unsupported query in the same multiplex from causing already-precomputed queries to be expanded again.

Benchmark

Measured on Ruby 4.0.0 using binary fragment expansion. The previous implementation was measured through an analyzer subclass that retains the original visitor path.

Levels	Query size	Before	After	Complexity	Speedup
10	573 bytes	11.832 ms	0.081 ms	3,071	146x
14	793 bytes	191.585 ms	0.092 ms	49,151	2,082x
16	903 bytes	823.619 ms	0.112 ms	196,607	7,354x
18	1,013 bytes	3,766.861 ms	0.117 ms	786,431	32,195x
20	1,123 bytes	timeout after 3,082 ms	0.125 ms	3,145,727	>24,000x

The level-20 query now completes complexity analysis instead of reaching the validation timeout.