Apparently `normalizer_normalize` is not verifying itself whether the string needs to be converted or not. Or does it at least not very performantly.

This simple change leads to a 4% performance gain on the processing of normalizeUnicode. Since this method is called quite often (i.e. for every file path) this has actually a measurable impact. For examples searches are now 200ms faster on my machine. Still not perfect but way to go.

Part of https://github.com/owncloud/core/issues/13221