Add automated benchmark pipeline

[RejectKid]

What changed

• add a dedicated BenchmarkDotNet GitHub Actions workflow
• run a fast Dry harness check across .NET 8, 9, and 10 on relevant pull requests
• run full cross-runtime measurements on relevant main pushes, every Monday, and by manual dispatch
• publish the Markdown report in the job summary and retain Markdown, HTML, CSV, and JSON artifacts
• document the pipeline and the limits of hosted-runner measurements in the README

Why

The benchmark suite was local-only. This makes it continuously executable and visible without treating diagnostic smoke timings or variable hosted-runner results as a strict regression gate.

Validation

• dotnet restore --locked-mode
• benchmark project Release build on .NET 10
• all 36 Dry-mode cases across .NET 8, 9, and 10
• confirmed Markdown, HTML, CSV, and JSON exports
• dotnet format --verify-no-changes --no-restore
• git diff --check