fix(pypi): find the repository in the README when project_urls has none

Selenium lists only www.selenium.dev in its metadata; its GitHub repository
appears only in the package description. A repository named after the package
is taken from there when no project link points at GitHub.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Author: fleizean

src/backend/PkgDiff.Infrastructure/Services/PyPIService.cs

@@ -163,9 +163,29 @@ private static bool IsGitHubUrl(string? url)
             .Concat(links.Where(l => !NotRepositoryKeys.Any(k => l.Key.Contains(k, StringComparison.OrdinalIgnoreCase)))
                 .Select(l => l.Value));
 
-        return candidates.Select(RepositoryRoot).FirstOrDefault(url => url != null);
+        return candidates.Select(RepositoryRoot).FirstOrDefault(url => url != null)
+            ?? RepositoryFromDescription(info);
     }
 
+    // Some packages link only their website (selenium: www.selenium.dev) and name the repository in the
+    // README alone. Only a repository named after the package counts there: READMEs link other projects too.
+    private static string? RepositoryFromDescription(PyPIPackageInfo info)
+    {
+        if (string.IsNullOrEmpty(info.Description) || string.IsNullOrEmpty(info.Name))
+            return null;
+
+        var name = NormalizeName(info.Name);
+        return System.Text.RegularExpressions.Regex.Matches(info.Description,
+                @"https?://(?:www\.)?github\.com/[\w.-]+/[\w.-]+",
+                System.Text.RegularExpressions.RegexOptions.IgnoreCase)
+            .Select(m => RepositoryRoot(m.Value.TrimEnd('.')))
+            .FirstOrDefault(url => url != null && NormalizeName(url[(url.LastIndexOf('/') + 1)..]) == name);
+    }
+
+    // PEP 503: names compare case-insensitively with runs of "-", "_" and "." as one "-"
+    private static string NormalizeName(string name) =>
+        System.Text.RegularExpressions.Regex.Replace(name, @"[-_.]+", "-").ToLowerInvariant();
+
     // https://github.com/{owner}/{repo} of a GitHub link (issues, a changelog file, releases), keeping a
     // /tree/{ref}/{folder} part that names the package folder of a monorepo; null for any other link
     private static string? RepositoryRoot(string? url)
@@ -245,6 +265,9 @@ private class PyPIPackageInfo
 
         [JsonPropertyName("project_urls")]
         public Dictionary<string, string>? ProjectUrls { get; set; }
+
+        [JsonPropertyName("description")]
+        public string? Description { get; set; }
     }
 
     private class PyPIReleaseFile