I don't know of any good summary of this, but I do recall heard accounts from specific speakers who say that they use something like [ɔ] before /l/ as a conditioned allophone of /ɑ/.
I would be very surprised if any speaker had a merger that applied in all environments except before /l/. It seems hard to prove that nobody has this, but I'm not familiar with any speaker who says they have this.
One thing that may be relevant is that even in accents that maintain a phonemic distinction between /ɑ/ and /ɔ/, British English /ɒl/ actually often corresponds to American English /ɔl/. I don't know of any good description of the exact contexts that conditioned this, but my impression is that for many non-cot-caught-merged American English speakers, /ɑl/ has a restricted distribution, occuring mainly or only before vowels. Judging from dictionary transcriptions, /ɔl/ is more-or-less standard (depending somewhat on the particular word) for non-cot-caught-merged American English speakers when "ol" is followed by a consonant or the end of a word (or in other words, in situations where the "l" is unambiguously a coda consonant). See my list of examples in What source explains the different pronunciations of "hol" in "alcohol" and "hollow"?
There is also a word spelled with "aul" where British English speakers all have short /ɒ/, but American English speakers may have either /ɔl/ or /ɑl/: cauliflower. Also, many British English speakers have /ɒl/ in words like vault, fault, false, spelled with "aul" or "al" followed by a voiceless consonant, but as far as I know non-cot-caught-merged American English speakers only have /ɔl/ in these words.
To sum up, I think your option (b) is more likely to be correct. It's true that speakers who don't distinguish caught and cot do maintain a distinction between north and start, but I don't think this is really analogous to a hypothetical merger of caught-cot with maintenance of a distinction between caller and collar (Peter Shor's example of a minimal pair). Rather, I would expect the pre-l context to be one of the first places where complete merger occurs, but it is quite likely that the merged vowel in this position will be realized as [ɔ], since even for non-merged speakers there is a tendency to use the phoneme /ɔ/ rather than /ɑ/ before /l/ in some contexts.
Short answer:
There's no doubt at all that the /t/ in relative may be flapped in standard American English. Indeed many dictionaries include a flapped /t/ in their transcriptions and give audio examples with flapped /t/'s too.
Here for example is the transcription from Cambridge Dictionaries:
relative
noun [ C ] UK /ˈrel.ə.tɪv/ US /ˈrel.ə.t̬ɪv/
That little upside-down hat under the 't' represents a flapped /t/. The audio there also uses a flapped /t/.
And here is the transcription from Oxford Dictionaries clearly indicating a flap for a /t/, where their convention is to represent flapped-/t/ with 'd':
relative /ˈrɛlədɪv/
In spite of the above, the /t/ in relative is only optionally flapped in this environment. Many speakers may not do so, and others my only use flapped /t/ in certain circumstances, for instance in connected speech as opposed to when giving a citation form. For why, see the full answer below.
Full answer:
Here is an excerpt from a paper by Alice Turk, Professor of Linguistic Phonetics at Edinburgh University, concerning the phonology of flaps in American English:
Here we are concerned with examples like (2) above, where a /t/ occurs between two unstressed vowels. Turk's example, provocative, has the same morphology as the Original Poster's word relative. In both instances the /t/ is intervocalic (occurs between two vowels), and occurs between unstressed vowels.
As described by Alice Turk, the flapping of /t/ in this environment is optional. This is well-attested in the comments here, where several respondents point out either that this /t/ is flapped, or that it can be.
In short then, [ɹɛləɾɪv] doesn't sound odd at all. Whether an individual happens to flap a /t/ in this sort of environment will depend on many factors: the personal habits and predilections of the speaker, whether they are saying the word alone or in a sentence, the speed at which they are talking, and so on and so forth.
Note:
No printed dictionaries give a transcription of relative with a secondary stress.
It has been suggested elsewhere here that the last syllable of relative has secondary stress. It doesn't, at least not in the normal meaning of the term.
In words with secondary stress, it's possible to have a rhythmic stress on both stressed syllables:
- 'abso-'lutely fan-'tastic.
In the utterance above we can have a full stress on each of the three bolded syllables, including both the first and third syllable of absolutely. We cannot do this kind of thing with the last syllable of relative
- 'famous 'rela-'tive (stress on -tive, badly formed).
Best Answer
From a phonological perspective, a typical transcription for an informal American English pronunciation of "she looked at me" would be:
This would probably be produced by a native American English speaker with a flap [ɾ] for the first /t/ and a glottal stop [ʔ] for the second /t/:
Both flap and glottal stop are allophones of /t/ in American English. The production of a /t/ phoneme as a flap [ɾ] does not mean that the underlying phonemic representation is a /d/—the phonemic representation /ʃiː lʊkd æt mi/ is not accurate.
The distinction between flap [ɾ] and [d] or [t] is whether or not there is pressure differential that builds up during tongue contact or not. The Wikipedia article about flap says "there is no buildup of air pressure behind the place of articulation and consequently no release burst."