Keep raw text in Search & Replace (nbsp and old mobi format)

For some reasons I want to convert to old mobi and I want to use non-breaking spaces. According to Amazon publisher guidelines it is not good in old mobi to use unicode   as non-breaking space, and I should use old and obsolete   named entity. I've checked this on my Paperwhite 4 and this is true - when there is unicode nbsp char, selection and highlighting mechanism threads that char, word before and word after as whole one word...

Problem is that when I try to use \1$nbsp; in calibre Search & Replace in conversion, Calibre will replace this with unicode nbsp char... I want to insert raw '$nbsp;' in output

How to force calibre to use raw text in replace pattern?