[markdown] AutolinkExtensionSyntax double-encodes percent-encoded sequences in the destination

[TsubasaYoshida]

When the GFM extended autolink syntax parses a bare URL that already contains percent-encoded sequences, the generated href gets double-encoded.

import 'package:markdown/markdown.dart';

void main() {
  print(markdownToHtml(
    'https://example.com/%40foo',
    extensionSet: ExtensionSet.gitHubFlavored,
  ));
}

Actual:

<p><a href="https://example.com/%2540foo">https://example.com/%40foo</a></p>

Expected (matches inline links and cmark-gfm's autolink output):

<p><a href="https://example.com/%40foo">https://example.com/%40foo</a></p>

The inline link form [text](https://example.com/%40foo) already produces the expected output, because normalizeLinkDestination leaves pre-existing percent-encoding alone per https://spec.commonmark.org/0.30/#example-502.

Cause: AutolinkExtensionSyntax.onMatch builds the href with Uri.encodeFull(destination), which escapes % itself, turning %40 into %2540. As a result, bare URLs whose path or query already contains percent-encoded characters link to a wrong destination.

I have a fix ready and will open a PR shortly.