build: add PublicApiAnalyzers baseline for both src packages (#96)

[Chris-Wolfgang]

Summary

Populates the PublicAPI.Shipped.txt / PublicAPI.Unshipped.txt baselines for the two shippable src/ packages so the PublicApiAnalyzers (RS0016 / RS0017 / RS0025) now guard the public API surface against unintended breaking changes — the goal of #96.

The analyzer infrastructure was already present on main: Directory.Build.props references Microsoft.CodeAnalysis.PublicApiAnalyzers 3.3.4 and globs the two .txt files as AdditionalFiles (opt-in per project via Exists()). What was missing was the actual baseline files in the two library projects. This supersedes the reverted A1 attempt (commits 7d2fe3a / 9e42df7, reverted in 38470e4 during v0.8.1 release prep).

What changed

• src/Wolfgang.Etl.TestKit/ — PublicAPI.Shipped.txt (full current public surface) + PublicAPI.Unshipped.txt (#nullable enable only).
• src/Wolfgang.Etl.TestKit.Xunit/ — PublicAPI.Shipped.txt (full current public surface) + PublicAPI.Unshipped.txt (#nullable enable only).
• Both .csproj files: <WarningsAsErrors>$(WarningsAsErrors);RS0016;RS0017;RS0025</WarningsAsErrors>.

Why the csproj WarningsAsErrors

In this repo's current analyzer configuration RS0016 ("symbol not in declared API") was not being raised, so a baseline alone would silently fail to catch additions to the surface (only RS0017 removals fired). Promoting the three tracking diagnostics to errors — scoped to the two opted-in src projects only, no protected-file edits — makes the baseline genuinely enforce both additions and removals. Verified by a negative test: deleting a single baseline entry now fails the build with RS0016.

Stale baseline corrected

The recoverable (reverted) baseline was generated without RS0016 active, so it omitted the recently-added contract-test methods (ExtractAsync_/TransformAsync_CurrentItemCount_tracks_each_yielded_item_Async, LoadAsync_/TransformAsync_stops_reading_source_at_MaximumItemCount_Async) and the implicit public constructors of all 16 contract base classes. Those entries were added so the baseline matches the real surface.

Public surface across TFMs

A single shared baseline builds clean against all 5 TFMs (net462;net481;netstandard2.0;net8.0;net10.0) for both projects — the public surface does not differ per TFM.

Build / test results

• dotnet build -c Release (all TFMs), both src projects: 0 warnings / 0 errors.
• dotnet test -c Release -f net8.0:
  • Wolfgang.Etl.TestKit.Tests.Unit: 58 passed
  • Wolfgang.Etl.TestKit.Xunit.Tests.Unit: 199 passed

Closes #96

🤖 Generated with Claude Code