Replace DiffPlex with a text diff API in DiffEngine

[SimonCropp]

Removes the DiffPlex dependency. DiffEngine now has its own public line diff, which the viewer uses, and which Verify can use to replace Verify.DiffPlex.

API

public static class TextDiff
{
    public static IReadOnlyList<DiffLine> Compute(string expected, string received);
    public static string Format(string expected, string received, TextDiffFormat format = TextDiffFormat.Compact);
}

public enum DiffLineKind { Unchanged, Removed, Added }
public record DiffLine(DiffLineKind Kind, string Text, int? ExpectedLine, int? ReceivedLine);
public enum TextDiffFormat { Full, Compact, Minimal }

• Compute returns lines in inline order: unchanged lines, and each changed block as its removed lines followed by its added lines.
• Format produces the three layouts Verify.DiffPlex has, including [BOF]/[EOF] and zero-padded line numbers in Compact.
• It compiles for every DiffEngine target, net462 included.

Implementation

• Myers O((N+M)D), linear-space middle-snake recursion.
• Works on spans: lines are ranges of the original strings, and equal lines get the same number from a span-keyed hash table, so no line is copied to compare it.
• Buffers come from the stack for small inputs and from ArrayPool otherwise.
• Compute makes a string only for each line it returns, and Format appends spans directly.

Differences from Verify.DiffPlex

• Whitespace and case count. Verify.DiffPlex ignored whitespace, so a failure caused only by whitespace showed no changed lines.
• Only trailing line breaks are trimmed from Format output, not trailing spaces, because a trailing space can now be the difference being reported.
• Line endings still don't count: the same lines under \r\n and \n compare as unchanged.

Packages and tests

• DiffPlex.dll is removed from every package: six copies in DiffEngine, two in the tray, one in each viewer package. The package listing snapshots are updated to match, and nothing else in them changed.
• Verify.DiffPlex is removed from the test projects.
• No existing viewer snapshot changed (renderer, scroll, minimal view and pixel).
• New TextDiffTests cover edge cases, all three formats and a 20,000-line input. A fuzz test checks 3,000 random cases against a brute-force answer, confirming every diff is as small as possible and rebuilds both texts exactly.
• New DiffRowsTests pin how the viewer pairs lines side by side.

Follow-up (Verify repos)

• Use TextDiff.Format as Verify's default string comparer.
• Deprecate Verify.DiffPlex.