fix: source generator benchmarks were running other generators

[TimothyMakkison]

Looks like when I'd run one benchmark it would run other source generators, giving a misleading time. For instance trying to benchmark StaticPropertyInitializationGenerator would also run TestMetadataGenerator and vice versa. You can prove this by changing all the predicates to false in TestMetadataGenerator, low and behold StaticPropertyInitializationGenerator would suddenly run much faster.

Strangely it doesn't happen to all generators, for instance InfrastructureGenerator only runs itself when benchmarked, making it harder for me to spot the issue 😠

Should solve the issue in #4639 / #4624 - I suspect I was consistently getting a faster time on the benchmarks because I was experimenting with TestMetadataGenerator, not expecting it to affect an unrelated source generator

StaticPropertyInitializationGenerator

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
RunGenerator	28.43 ms	0.542 ms	1.201 ms	666.6667	333.3333	8 MB

TestMetadataGenerator

Method	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
RunGenerator	201.6 ms	12.64 ms	36.28 ms	191.5 ms	10000.0000	4000.0000	97.07 MB

AotConverter

Method	Mean	Error	StdDev	Gen0	Allocated
RunGenerator	60.76 ms	1.212 ms	2.761 ms	1000.0000	9.29 MB

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

---

This is a clean, focused fix that correctly addresses the benchmark pollution issue. The single-line change using WithAnalyzerReferences([]) ensures that only the explicitly benchmarked generator runs, not all generators referenced by the test project. The code:

• Uses modern C# 12 collection expression syntax
• Correctly applies the Roslyn API to clear analyzer references
• Aligns with TUnit's "Performance First" principle by improving benchmark accuracy
• Has no security or logic concerns