Do not parse hard line breaks when doing link label blocks

Consider this markdown:

```markdown
[x\

]: https://rust-lang.org
```

That's not a link definition, but pulldown-cmark thought it was.
The hard line break on line 1 is an inline element, and should not
prevent the paragraph break from parsing.

specs/regression.txt

@@ -1560,3 +1560,14 @@ ISSUE xxx
 </li>
 </ul>
 ````````````````````````````````
+
+ISSUE 753
+
+```````````````````````````````` example
+[x\
+
+]: https://rust-lang.org
+.
+<p>[x\</p>
+<p>]: https://rust-lang.org</p>
+````````````````````````````````

src/linklabel.rs

@@ -22,7 +22,7 @@
 
 use unicase::UniCase;
 
-use crate::scanners::{is_ascii_whitespace, scan_eol};
+use crate::scanners::{is_ascii_whitespace, scan_eol, is_ascii_punctuation};
 use crate::strings::CowStr;
 
 #[derive(Debug)]
@@ -60,7 +60,7 @@ pub(crate) fn scan_link_label_rest<'t>(
         match *bytes.get(ix)? {
             b'[' => return None,
             b']' => break,
-            b'\\' => {
+            b'\\' if is_ascii_punctuation(*bytes.get(ix + 1)?) => {
                 ix += 2;
                 codepoints += 2;
                 only_white_space = false;

tests/suite/regression.rs

@@ -1838,3 +1838,16 @@ fn regression_test_118() {
 
     test_markdown_html(original, expected, false, false, false);
 }
+
+#[test]
+fn regression_test_119() {
+    let original = r##"[x\
+
+]: https://rust-lang.org
+"##;
+    let expected = r##"<p>[x\</p>
+<p>]: https://rust-lang.org</p>
+"##;
+
+    test_markdown_html(original, expected, false, false, false);
+}