build: add BenchmarkDotNet baseline project for test doubles (#80, #102)

[Chris-Wolfgang]

Summary

Stands up a BenchmarkDotNet baseline project for the ETL Test Kit's runtime test doubles, satisfying #80 (stand up a BDN baseline project) and #102 (hold benchmarks to TreatWarningsAsErrors via a curated benchmarks/.editorconfig, with no benchmarks/Directory.Build.props exemption).

The project

benchmarks/Wolfgang.Etl.TestKit.Benchmarks/ — an Exe targeting net10.0, LangVersion=latest, ImplicitUsings=disable, with a ProjectReference to Wolfgang.Etl.TestKit and a single BenchmarkDotNet 0.15.8 package reference. Program.cs uses the standard BenchmarkSwitcher.FromAssembly(...).Run(args) entry point. Analyzer package references are inherited from the root Directory.Build.props (comment noted in the csproj). Added to ETL-Test-Kit.slnx under the /benchmarks/ folder.

Three benchmark classes

Each is [MemoryDiagnoser], with [Params(1_000, 10_000, 100_000)] public int ItemCount and a [GlobalSetup] that builds an in-memory int[]. T = int keeps allocations dominated by the framework path rather than record construction. One representative benchmark per class is [Benchmark(Baseline = true)].

• ExtractorBenchmarks.Extract — new TestExtractor<int>(int[]) then await foreach over ExtractAsync().
• LoaderBenchmarks.Load — new TestLoader<int>(collectItems: false) then await loader.LoadAsync(source).
• TransformerBenchmarks.Transform — new TestTransformer<int>() then await foreach over TransformAsync(source).

#102 — held to TreatWarningsAsErrors

There is no benchmarks/Directory.Build.props. The benchmark project inherits the root Directory.Build.props, so it IS held to TreatWarningsAsErrors in Release. The only relaxation is a curated benchmarks/.editorconfig silencing genuinely benchmark-inappropriate analyzer rules, each with a one-line justification:

• AsyncFixer01 — remove async/await; benchmark methods follow BDN conventions.
• MA0004 — ConfigureAwait(false) not needed in the benchmark harness.
• S1215 — GC.GetTotalMemory is required by memory benchmarks.
• VSTHRD200 — Async suffix; benchmark method names follow BDN naming conventions.

This matches the curated FixedWidth set exactly — no additional rule IDs were needed; the Release build is clean (0 warnings, 0 errors) against the full analyzer set.

IAsyncEnumerable source

Rather than depend on System.Linq.Async's .ToAsyncEnumerable() (a transitive dep), the loader and transformer benchmarks use a tiny local async IAsyncEnumerable<int> generator helper, keeping the source self-contained and the measured pipeline obvious.

Verification

• dotnet build benchmarks/Wolfgang.Etl.TestKit.Benchmarks/... -c Release → Build succeeded, 0 Warning(s), 0 Error(s).
• dotnet run -c Release --project benchmarks/... -- --list flat lists all three benchmarks.
• dotnet run -c Release --project benchmarks/... -- --filter '*Extract*' --job dry executes and emits a summary table with the MemoryDiagnoser Allocated column (e.g. ~520 B/op across all three ItemCount params), confirming the harness runs.

Closes #80
Closes #102

🤖 Generated with Claude Code