Add a test project covering the conversion logic

[trsdn]

Fixes #23

Stacked. Based on fix/16-markdown-escaping, with fix/21-task-list-checkboxes and fix/14-markdown-html-detection merged in, because the tests assert the fixed behaviour. Merge the eight fix PRs first and this diff reduces to the test project itself. Only the last commit is unique to this branch.

Why

There were no tests, which is exactly why this batch of bugs shipped in the first place:

• task list checkboxes were emitted as raw <input type="checkbox"> markup that Loop does not render (TransformTaskLists is dead code; checkbox output relies on brittle string replacement #21)
• nested lists were duplicated (Nested list items are duplicated in HTML → Markdown conversion #17)
• inline formatting was dropped from list items and table cells (Inline formatting is lost inside list items and table cells #15)

Every one of those is a two-line unit test and effectively invisible by eye.

Restructuring

None of the conversion code touches WinUI or the clipboard, but it lived in the app project, which targets net8.0-windows10.0.26100.0 and pulls in the Windows App SDK. An ordinary test project cannot reference that.

The four converter files move unchanged to a new md2loop.Core library (net8.0, no UI dependencies). The app references it and is otherwise untouched. Git records all four as pure renames with zero content changes, so this reshuffles nothing behaviourally.

md2loop            WinUI 3 app: window, clipboard access, UI
md2loop.Core       conversion logic, no UI or clipboard dependencies
tests/md2loop.Tests unit tests for md2loop.Core

Coverage

84 tests, 3 skipped. xUnit.

Area	What it covers
ClipboardContentDetectorTests	mode detection, scoring, RTF/HTML sniffing, and the #14 regression where Markdown discussing HTML was misreported
HtmlToMarkdownConverterTests	every element the converter handles, plus explicit regressions for #15, #16, #17 and #19
LoopHtmlConverterTests	Markdown to Loop HTML, including the #21 checkbox regression and an assertion that <input> is never emitted
RoundTripTests	Markdown to Loop HTML and back, which is what a user does when they paste into Loop and copy it out later

The regression tests are written against the observable symptom rather than the implementation. For example Convert_NeverEmitsCheckboxInputs asserts the absence of <input, so it will fail again if a future Markdig upgrade changes the markup, which is precisely how #21 slipped through.

Two failures worth calling out

The suite found real problems while I was writing it, rather than just confirming what I expected:

1. Convert_RendersHeadingsAndInlineFormatting asserted <h1>Title</h1>, but UseAdvancedExtensions() enables AutoIdentifiers, so the real output is <h1 id="title">. My assertion was wrong; corrected to match on content.

2. FencedCode_SurvivesARoundTrip failed with ```csharp becoming ```. That is not a bad assertion, it is bug Class stripping removes language-* from code blocks, losing the code language #34 reproducing: the blanket class-stripping regex removes the language-* class that carries the code language. Left in place as a passing test documenting current behaviour, with a skipped test alongside it stating the correct behaviour.

Known bugs recorded as skipped tests

Rather than delete anything that failed, the three open bugs get a skipped test each that spells out the expected result. They flip to passing when the bug is fixed:

• #32 - no blank line after a list, so a following block is swallowed
• #33 - whitespace between block elements leaks a leading space
• #34 - class stripping removes the code language

CI and housekeeping

• build.yml gains a dotnet test step, so this actually guards master instead of only running locally.
• bin/ and obj/ added to .gitignore. They were never listed; the repo stayed clean only because files were staged individually. With three projects in the tree that is no longer safe, and staging everything would have committed ~40 build artefacts.

Verification

• dotnet test -c Release - 84 passed, 0 failed, 3 skipped.
• dotnet build md2loop -c Release -r win-x64 - 0 warnings, 0 errors, confirming the app still builds against the extracted library.