Optimize repeated FieldsWillMerge comparisons

[ydah]

Improve the performance of FieldsWillMerge validation for queries containing many repeated composite fields and shared fragment selections.

Previously, the validator could repeatedly compare the same expanded fragment fields for every pair of parent fields. This caused nested Cartesian comparisons and severe superlinear runtime growth before query execution or authentication.

This change:

• groups fields with identical sub-selections and compares only one representative from each group
• memoizes comparisons between equivalent expanded field groups
• keeps mutually exclusive and non-exclusive comparisons separate
• adds a regression test that verifies a divergent conflict is still detected without reaching the default validation timeout

Benchmark

Measured using a query with repeated composite fields and a shared fragment containing repeated fields.

Fields	Query size	Before	After	Speedup
100	2,320 bytes	193.6 ms	3.14 ms	61.7x
200	4,552 bytes	1,438.4 ms	3.72 ms	386.7x
400	9,000 bytes	11,732.5 ms	8.68 ms	1,351.7x
800	17,864 bytes	did not finish within 30 seconds	20.08 ms	—

The 400-field case is reduced by approximately 99.93%.

[swalkinshaw]

Looks like a good optimization to me

[rmosolgo]

Thanks for taking a look, @swalkinshaw . The only thing that stood out to me was the use of to_query_string, which creates a big string for large nested selections. But I think it's pretty optimized in terms of object allocations, and the speedup here is worth it. Thanks for sharing this, @ydah!