To summarize the findings so far:
StackExchange does insert unicode characters U+200C (zero width non joiner) and U+200B (zero width space) into comments on occasion. Link to Meta.SE post. This is done for reasons related to how browsers do line breaking.
In the linked Meta.SE thread it is stated that this is done only when there is a run of at least 80 non-whitespace characters. However, in the examples shown under the post above, we don't have such a long run of non-whitespace characters! In @Kuba's comment the longest run is 35.
This makes me suspect that there is a recent change in how comments are handled. Perhaps a bug? It would be good to have an answer from the SE staff.
Incidentally, there was a question today about a non-working command which is broken exactly due to an invisible zero-width space. The OP claims that he didn't copy any part of that command from a comment, but the coincidence in timing makes me wonder ...