FileRW list sortiert und gedeckelt (#673), Taskboard task_find (#674)

Die Betriebskontrolle im Testbetrieb meldete Stundendateien als fehlend, die
es gab: 105 Eintraege in der Reihenfolge des Dateisystems, vierzeilig als
eingeruecktes JSON. list liefert jetzt die 50 zuletzt geaenderten Eintraege,
neueste zuerst, eine Zeile je Eintrag, mit Gesamtzahl im Kopf; pattern, sort,
order und limit steuern das.

Dabei aufgefallen (#675): Directory.GetFiles wertet Trennzeichen im Muster
aus - pattern "../*" fuehrte an der Pfadpruefung vorbei aus dem Verzeichnis
heraus, in list wie in read_many. Muster mit '/', '\' oder '..' werden jetzt
abgewiesen.

Und derselbe Befund ist keine zweite Aufgabe (#674): task_find beantwortet
die Frage vor dem Anlegen, task_create legt bei gleichbedeutendem Titel einer
offenen Aufgabe keine zweite an, sondern vermerkt das erneute Auftreten an
der bestehenden. force=true geht trotzdem, Erledigtes blockiert nichts.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

src/ClawdDotNet.Core/Tasks/TaskSimilarity.cs

@@ -0,0 +1,125 @@
+using System.Globalization;
+using System.Text;
+
+namespace ClawdDotNet.Core.Tasks;
+
+/// <summary>How closely two task titles describe the same thing.</summary>
+public enum TaskTitleMatch
+{
+    /// <summary>Different subjects.</summary>
+    None,
+
+    /// <summary>Related enough to be worth showing, not close enough to refuse a new task.</summary>
+    Similar,
+
+    /// <summary>The same finding said twice.</summary>
+    Duplicate
+}
+
+/// <summary>
+/// Title comparison for the taskboard (#674).
+///
+/// <para>Observed in the test operation: an agent that runs every three hours re-reported the
+/// same gap on every run. Six tasks said "Eur Kurse nicht aktuell", five said the news were
+/// missing. Nobody can tell from that whether a problem is new or three days old.</para>
+///
+/// <para>The comparison is deliberately shallow - normalise, then compare word sets. Two titles
+/// that differ only in word order, case, umlaut spelling or punctuation are the same title;
+/// a title fully contained in a longer one is the same finding with more detail. Anything
+/// cleverer would be guesswork a cheap model cannot audit.</para>
+/// </summary>
+public static class TaskSimilarity
+{
+    /// <summary>Jaccard overlap from which two titles count as the same finding.</summary>
+    public const double DuplicateThreshold = 0.8;
+
+    /// <summary>Overlap from which two titles are worth showing side by side.</summary>
+    public const double SimilarThreshold = 0.5;
+
+    /// <summary>Statuses that still ask for attention - only these can be duplicated.</summary>
+    public static bool IsOpen(TaskItemStatus status) => status is
+        TaskItemStatus.Backlog or TaskItemStatus.Todo or TaskItemStatus.InProgress
+        or TaskItemStatus.InReview or TaskItemStatus.Blocked;
+
+    /// <summary>Lower case, umlauts folded, punctuation dropped, whitespace collapsed.</summary>
+    public static string Normalize(string? title)
+    {
+        if (string.IsNullOrWhiteSpace(title)) return "";
+
+        var folded = new StringBuilder(title.Length);
+        foreach (var c in title.ToLowerInvariant())
+        {
+            switch (c)
+            {
+                case 'ä': folded.Append("ae"); break;
+                case 'ö': folded.Append("oe"); break;
+                case 'ü': folded.Append("ue"); break;
+                case 'ß': folded.Append("ss"); break;
+                default: folded.Append(c); break;
+            }
+        }
+
+        // Remaining diacritics (é, ç, …) lose their marks rather than their letter.
+        var decomposed = folded.ToString().Normalize(NormalizationForm.FormD);
+        var plain = new StringBuilder(decomposed.Length);
+        foreach (var c in decomposed)
+        {
+            if (CharUnicodeInfo.GetUnicodeCategory(c) == UnicodeCategory.NonSpacingMark) continue;
+            plain.Append(char.IsLetterOrDigit(c) ? c : ' ');
+        }
+
+        return string.Join(' ', plain.ToString().Split(' ', StringSplitOptions.RemoveEmptyEntries));
+    }
+
+    /// <summary>Word set of a normalised title.</summary>
+    private static HashSet<string> Words(string normalized)
+        => new(normalized.Split(' ', StringSplitOptions.RemoveEmptyEntries), StringComparer.Ordinal);
+
+    /// <summary>How closely two titles match.</summary>
+    public static TaskTitleMatch Compare(string? left, string? right)
+    {
+        var a = Normalize(left);
+        var b = Normalize(right);
+        if (a.Length == 0 || b.Length == 0) return TaskTitleMatch.None;
+        if (a == b) return TaskTitleMatch.Duplicate;
+
+        var wordsA = Words(a);
+        var wordsB = Words(b);
+        var shared = wordsA.Intersect(wordsB, StringComparer.Ordinal).Count();
+        if (shared == 0) return TaskTitleMatch.None;
+
+        // "Betriebsluecke: Eur Kurse" inside "Betriebsluecke: Eur Kurse nicht aktuell" is the
+        // same finding, told at different lengths. A single shared word is not - "Wetter"
+        // would swallow every weather task there is.
+        var (smaller, larger) = wordsA.Count <= wordsB.Count ? (wordsA, wordsB) : (wordsB, wordsA);
+        if (smaller.Count >= 2 && shared == smaller.Count) return TaskTitleMatch.Duplicate;
+
+        var overlap = (double)shared / (wordsA.Count + wordsB.Count - shared);
+        if (overlap >= DuplicateThreshold) return TaskTitleMatch.Duplicate;
+        if (overlap >= SimilarThreshold) return TaskTitleMatch.Similar;
+        return TaskTitleMatch.None;
+    }
+
+    /// <summary>
+    /// Open tasks whose title describes the same thing as <paramref name="title"/>,
+    /// closest match first. <paramref name="exceptId"/> skips the task being edited.
+    /// </summary>
+    public static List<(TaskItem Task, TaskTitleMatch Match)> FindOpen(
+        IEnumerable<TaskItem> tasks, string? title, string? exceptId = null)
+    {
+        var hits = new List<(TaskItem Task, TaskTitleMatch Match)>();
+        foreach (var task in tasks)
+        {
+            if (!IsOpen(task.Status)) continue;
+            if (exceptId is not null && string.Equals(task.Id, exceptId, StringComparison.OrdinalIgnoreCase)) continue;
+
+            var match = Compare(task.Title, title);
+            if (match != TaskTitleMatch.None) hits.Add((task, match));
+        }
+
+        return hits
+            .OrderByDescending(h => h.Match)
+            .ThenByDescending(h => h.Task.UpdatedAt)
+            .ToList();
+    }
+}