I supposed you could argue that the bug itself was a hard fork, since it allowed the attacker to do something which everyone agreed should be invalid (create 92 billion bitcoin out of thin air). But the successful patch to fix it was a soft fork, as it was simply tightening the rules (by making such overflow incidents invalid). Even BIP50 wasn't really a hard fork, despite often being called such. I don't believe bitcoin has ever had a successful hard fork.

The presence or absence of a chain split does not give an indication as to whether a change was a soft or a hard fork. Although obviously a hard fork could result in a chain split, we could also code a hard fork (such as to fix the 2106 problem) decades in advance which would result in everybody running a compatible client well in advance and there being no chain split.