Better workaround for Windows encoding issue.

Aaron Crane did some testing, and found that the error can be suppressed by
any of:

- removing the :crlf layer
- using the :utf8 layer instead of :encoding(UTF-8)
- using a file that ends in \r\n rather than \n
- removing the C<< eof $fh >> test
- leaving $/ unmodified, rather than setting it to undef

So I decided to remove the $/. Removes a line of code, too.

Changes

@@ -1,8 +1,9 @@
 Revision history for Test-File-Contents
 
 0.23
-      - Fix Windows encoding conversion by reading files in raw bytes before
-        converting to the encoding.
+      - Fix encoding errors when reading in a file on Windows with `utf8`
+        enabled. See https://rt.perl.org/Ticket/Display.html?id=127668 for the
+        bug we've worked around.
 
 0.22  2016-03-02T23:54:26Z
       - Adjusted tests to account for change in output of Text::Diff 1.44.

lib/Test/File/Contents.pm

@@ -436,11 +436,12 @@ sub _slurp {
     my ($file, $encoding) = @_;
     my $layer = !$encoding  ? ''
         : $encoding =~ '^:' ? $encoding
-        :                     ":raw:encoding($encoding)";
+        :                     ":encoding($encoding)";
     open my $fh, "<$layer", $file or return;
     return '' if eof $fh;
-    local $/;
-    return <$fh>;
+    # Don't use `local $/; return <$fh>;`, it does not work on Windows.
+    # See https://rt.perl.org/Ticket/Display.html?id=127668 for details.
+    return join '', <$fh>;
 }
 
 sub _resolve {