As varro says, it's always long. This is predictable as one example of a more general rule that any vowel is always long before "ns" or "nf".
This rule of vowel length is thought to be the result of "compensatory lengthening" or assimilation. We have evidence that the consonant sound [n] could be lost before [s] in Latin: thus the suffix -ensis corrresponds to Italian -ese, Portuguese -ês, and French -ois/-ais. In Latin, the loss of [n] in the sequence [ns] shows up sporadically, and different words show different frequencies of [n]-loss. For example, the adjective endings -ēnsis and -ōsus both originally had -ns-, but you'll probably be more likely to encounter spellings with n for the first and without n for the second when reading Latin texts.
The loss of [n] in words like these isn't thought to have happened all in one step. The nasal consonant is supposed to have fused with the preceding vowel, producing a long nasalized vowel: something like [ens] > [ẽːs]. Then later on, this long nasalized vowel lost its nasality, resulting in the same outcome as a non-nasal long vowel [eː] in the modern Romance pronunciations of the suffixes descended from -ēnsis.
(I don't know enough about phonetics to give a detailed explanation, but the loss of [n] before the voiceless fricative [s] and the corresponding "compensatory lengthening" of the preceding vowel is a fairly phonetically natural sound change. A similar change applied in the history of English, and is why our word goose/geese corresponds to Latin anser.)
W. Sidney Allen (1978) actually says that in Classical Latin, the nasal consonant was restored in the pronunciation of words spelled with ns (at least in some words and in higher registers of speech), but due to the "artificial" nature of this restoration, the vowel continued to be pronounced long, resulting in what some call a "super-heavy" syllable: a syllable with a long vowel followed by a coda consonant. In other words, according to Allen, -ēnsis was pronounced in the classical period as [eːnsɪs]. (In Latin, "super-heavy" syllables behave exactly like normal heavy syllables metrically, as varro mentions.)
The development in a word such as consul, therefore, is: prehistoric cŏnsol, early Latin cō̃sol; classical colloquial cōsul; classical literary cōnsul. (Vox Latina 2nd edition, p. 29-30)
The consonant [f], which is a voiceless fricative like [s], seems to have caused the same kind of lengthening in a sequence of a vowel followed by /nf/. However, unlike with ns, there isn't evidence of nasality being completely lost in words with nf. The absence of denasalized pronunciations can be explained by the fact that the sequence -nf occurred almost exclusively in words where the n and f could be identified as belonging to separate morphemes, such as words with the prefix in- or con-. Analogy with words where these prefixes came before another consonant would have strengthened the tendency to use a pronunciation with a long vowel followed by restored consonantal /n/ rather than a denasalized long vowel.
Unmarked doesn't necesssarily mean "short"
Lewis & Short/Elem. Lewis don't always mark the length of a vowel. If it has a breve, they're saying it's short, but if it's unmarked, they aren't saying either way: you don't know if it is long or short. Sometimes this could be because they were unsure, or sometimes it's because they don't think the information is important (for example, vowel length doesn't affect scansion in closed syllables, and the dictionary assumes that the reader will already be familar with the length of vowels in common suffixes such as -us, -i).
If a vowel letter is marked with a macron in L&S, that usually means it corresponds to a long vowel, but there are a few questionable cases like
"lucubro", which L&S give as lūcū^bro but which probably always had a short "u" in the second syllable. The second syllable would be able to scan long even with a short vowel because of variable syllable division of "br" as ".br" or "b.r". Wiktionary gives the etymology as Proto-Indo-European *lewk-o-dʰro-, which wouldn't regularly give rise to a long vowel before the br.
"horizon," which L&S give as "hŏrīzon" but which probably had a short "i", in a syllable that scanned long because of the double consonant "z"
"major," which L&S give as "mājor" but which probably had a short "a", in a syllable that scanned long because of the double consonant "j"
Overall, I'd recommend being a bit suspicious of macrons in L&S on vowels in the following contexts: before a mute + liquid cluster, before "z" or "j", and occasionally before a word-final consonant (my impression is that it's somewhat unclear whether words like far, farris ever had a long vowel in the nominative singular; some authors suggest that instead the consonant was long before a following word starting with a vowel). I give a few more references and examples in my answer to What effect should a macron have on the sound of a letter and its word?
Wiktionary does not use breves, so in that specific context unmarked vowel letters are supposed to be short (or non-syllabic).
The only inconsistency I see in the dictionary entries that you cite is the lack of a macron in Wiktionary's entry for "Carthāginiensis", which is presumably just an error.