Apparently the understanding of this has gotten considerably more elastic since the last time I checked in.
I have always understood it as I described it, and I've heard
others understand it that way too. It's described this way on the
KAG's
wiki, for instance. Lawrence's question was pitched
specifically to address the ship-in-which-I-fled problem.
What I understood from Okrand's statement was that the head noun of the relative clause (which could be the subject or the object) can only be the subject or object of the main verb as well, due to the intrinsic nature of Klingon. That is, subjects and objects in Klingon are unmarked, and that holds true for nouns in a relative phrase, also.
Except they don't have to be unmarked. Leaving aside -'e', which I've said doesn't really act like a syntactic suffix when it's used as emphasis, it's perfectly legal, though redundant, to use a locative-marked noun as the object of a verb whose meaning is inherently locative, as in the canonical sentence bIQtIqDaq vIjaH for I'm going to the river.
Now, there probably aren't any verbs that let you use -mo'
or -vaD as subject or object, but this is because no verbs
have arguments that are inherently causes or beneficiaries, so far
as we know, not because putting them on verb arguments is illegal
per se.
Any other nouns in the phrase (not in a N1-N2 construction or a timestamp) require a suffix and are therefore marked. For either of those nouns in a relative phrase to take a Type-5 suffix, they would simultaneously be both marked and unmarked. I don't think Okrand could figure out how that would work, and neither can I.
Believe me, I have long and fervently wished you could use Type-5s on nouns in a relative phrase (and probably did it a time or two), but I no longer think it is possible.
Your last example just reads to me "I am on a ship spying on (something) I am evading." I don't believe that {Duj} can simultaneously be a locative with {-Daq} and the object of {jun}. Is there canon that supports this?
We have yIntaHbogh nuvpu'Daq HItlhej Come with me to the living in paq'batlh. The subject of yIntaHbogh is marked as a locative.
paq'batlh also gives us 'ej Hoch vengHomDaq Hoch vengDaq je / Suchbogh ghaH qeylIS luQoy And Kahless spoke to them / In every village and city he went. In this one the object of Suchbogh is a double locative noun-noun, Hoch vengHomDaq Hoch vengDaq je. Those are not locatives of the verb Such, because Kahless did not visit people in general while in those places; he visited those places. I could see, however, how this one might be a little more ambiguous.
It has chalqachlIj rachlu'ta'bogh tutDaq / mol'egh betleH The bat'leth sunk into the post / Of your fortified tower. tutDaq is the subject and head noun of the relative clause, which is itself a locative of the main clause.
Skybox 99 has qIb HeHDaq, 'u' SepmeyDaq Sovbe'lu'bogh lenglu'meH He ghoSlu'bogh retlhDaq 'oHtaH. The relative clause 'u' SepmeyDaq Sovbe'lu'bogh unknown regions of the universe has 'u' SepmeyDaq as its object, and it's a locative of the main clause.
TKW has loghDaq Suvrupbogh SuvwI'pu' chaH Hoch SuvwI'pu''e' In space, all warriors are cold warriors. loghDaq in space is the object of Suvrupbogh who are ready to fight, and the relative clause is a locative of the main clause.
I haven't done an exhaustive search, but that's already quite a lot of evidence, and some of it is quite old.
vaj yIlop! SaH pab DaneHbogh!
-- SuStel http://trimboli.name