fix emoji size when using "marked" markdown

The old logic of determing whether a message consists solely of emoji
was pretty primitive: it just checked whether there are top level text
nodes that contain text.

This fails when using the "marked" markdown parser because it can return
a <p> node, or a <table> node or whatever as the first element; the
old logic would conclude that the message only uses emoji and render
the emojis in "big".

Rewrite the logic to scan every child node, ignoring "emoji" nodes, and
when there is a text node with some text, we know that the message
doesn't consist solely of emojis.

This approach also works with the "original" markdown parser.

app/emoji/client/emojiParser.js

@@ -32,7 +32,30 @@ Tracker.autorun(() => {
 
 			const emojis = Array.from(checkEmojiOnly.querySelectorAll('.emoji:not(:empty), .emojione:not(:empty)'));
 
-			const emojiOnly = emojis.length && !Array.from(checkEmojiOnly.childNodes).filter((node) => node.nodeType === Node.TEXT_NODE).map((el) => el.nodeValue).join('').trim();
+			const walker = document.createTreeWalker(
+				checkEmojiOnly,
+				NodeFilter.SHOW_ELEMENT | NodeFilter.SHOW_TEXT,
+				(node) => {
+					if (node.nodeType === Node.ELEMENT_NODE && (
+						node.classList.contains('emojione')
+						|| node.classList.contains('emoji')
+					)) {
+						return NodeFilter.FILTER_REJECT;
+					} // else
+					return NodeFilter.FILTER_ACCEPT;
+				}
+			);
+
+			let hasText = false;
+
+			while (walker.nextNode()) {
+				if (walker.currentNode.nodeType === Node.TEXT_NODE && walker.currentNode.nodeValue.trim() !== '') {
+					hasText = true;
+					break;
+				}
+			}
+
+			const emojiOnly = emojis.length && !hasText;
 
 			if (emojiOnly) {
 				for (let i = 0, len = emojis.length; i < len; i++) {