Perf Phase 0: benchmark baseline for hot analyzers

[credfeto]

Part of #455.

The benchmark project currently covers only SuppressMessageDiagnosticsAnalyzer (FunFair.CodeAnalysis.Benchmark.Tests/Bench/SuppressMessageAnalyzerBenchmark.cs, with allocation-ceiling assertions in SuppressMessageAnalyzerBenchmarkTests.cs). Before optimizing anything, add benchmarks in the same pattern for the analyzers targeted by later phases, so each phase has a measurable before/after.

• Benchmark for ProhibitedMethodsDiagnosticsAnalyzer (fires on every SimpleMemberAccessExpression — the hottest path in the package)
• Benchmark covering the invocation-analyzer family (ProhibitedMethodInvocations, ForceMethodParametersInvocations, ProhibitedMethodWithStrictParametersInvocation)
• Benchmark for the TestClass* trio / TestDetection.IsDerivedFromTestBase paths
• Record baseline numbers (mean time + allocated bytes) in this issue before Phase 1 starts

Verification: new benchmark classes run cleanly via the existing dotnet test-driven benchmark harness (see SuppressMessageAnalyzerBenchmarkTests.cs for the pattern of running the benchmark and asserting allocation ceilings).