I assume your question is not why the author of a grammar would state a fact about a language when no known languages act contrarily, it is whether any languages act contrarily. I don't know why you expect the Greek case (I do as well, I thought I'd tell you why Greek is expected, also why that basis is not invariant across languages).
The general rule is that a nominal modifier agrees with the head of its phrase. At the lowest level, that would mean that adjectives agree with the head noun in the NP, then determiners will agree in gender with the NP and verbs will agree with the DP. Sometimes there are conflicting genders (when there are multiple nominals with different features), for example "the cat and the cow" where the entire conjunction can't have both cat-gender and cow-gender, therefore the conflict must be resolved. Or, "The cat / cow is a monster". In the latter case the general rule is that the verb (if any) agrees with the subject rather than the object.
Greek reflects this typical logic for resolving agreement conflicts, but not all languages do. Shona has 19 genders divided into formal vs. semantic groups, where formal genders are those that are lexically fixed and semantic one are non-lexical and reflect derived semantics (locatives, augmentatives and diminutives). Modifiers agree with the head noun, thus chi-toro chi-kuru 'big store' (lit. 'store big'), zvi-toro zvi-kuru 'big stores', mw-ana mu-kuru 'big child'. The head can have two genders, one semantic locative and the other formal lexical: adjectives can agree with either (ku-chi-toro chi-kuru or ku-chi-toro ku-kuru 'to the big store'). There are various ways to make theoretical sense of this, for example either attach the adjective under the lower NP or under the higher locative PP; or make the lower noun gender optionally invisible or the higher locative gender transparent. There is a simple explanation, but it isn't self-evident how you get these two patterns, therefore the grammarian is obligated to explicitly recognize the facts.
Relative pronouns agree in gender with the clause internal gap which is the head, as in chi-toro cha ndakaona __ 'the store which I saw' (lit. 'store which I-saw'). Here, we cannot tell if the relative complementizer agrees with the overt higher head chitoro or the lower trace, because the features are the same. We expect theoretically that agreement is with the overt noun chitoro because "which I saw" is a modifier of "store".
The internal head can also be locative, since you can say ndakaenda ku-chitoro 'I went to the store', and agreement can be with the lexical class. When the nominal is locative in the lower clause (not the higher clause), there are still two patterns of agreement on the higher relative pronoun: lexical (vakaona chi-toro cha ndakaenda __ 'they saw the store which I went to' lit. 'they-saw store which I-went'), and semantic
(vakaona chi-toro kwa ndakaenda __ lit. 'they-saw store to-which I-went').
The puzzle here is, why does the higher relative complementizer kwa- optionally agree with the "upper" gender which disagrees with the overt head's gender, and does not mandatorily agree with the overt N's lexical gender? This may cause some theoretical distress, but descriptively speaking the generalization is clear: relative pronouns can agree formally or semantically, and sometimes that gives different agreement patterns.
Linguists who have a modicum of knowledge of crosslinguistic variability in how agreement words will therefore tend to explicitly state the ordinary case, when they know that the ordinary case is not mandatory in all languages.