Fixed regression in parsing of HTML comments in markdown...

and other non-HTML formats (`Text.Pandoc.Readers.HTML.htmlTag`).
The parser stopped at the first `>` character, even if it wasn't
the end of the comment.

Closes #4019.

src/Text/Pandoc/Readers/HTML.hs

@@ -1172,8 +1172,9 @@ htmlTag f = try $ do
   case next of
        TagComment s
          | "<!--" `isPrefixOf` inp -> do
-          char '<'
-          manyTill anyChar endAngle
+          string "<!--"
+          count (length s) anyChar
+          string "-->"
           stripComments <- getOption readerStripComments
           if stripComments
              then return (next, "")

test/command/4019.md

@@ -0,0 +1,8 @@
+```
+pandoc --wrap=preserve
+This <!-- x > 0 --> works!
+This <!-- x > 0 --> fails?
+^D
+<p>This <!-- x > 0 --> works!
+This <!-- x > 0 --> fails?</p>
+```