The pool is small enough that the concept of "consensus" is mildly anomalous, but I think the consensus would be that we can't tell what the proto vowel system was phonetically. The standard view is that PB had 3 vowel heights. The first degree of height may be graphically represented as [i u, î û, i̧ u̧] (cedillas); the second is [ɪ ʊ, i u, e o] (although i u is not used if the highest vowels are written as i u); and the lowest vowels are written [e o, ɛ ɔ], again obviously not e o if the second degree of vowel is written as e o. Schadeberg's paper in The Bantu Languages has a table comparing various systems of notation. More to the point, these are just ways of writing the contrasts, and one should not over-interpret use of symbols as constituting phonetic claims: there are three vowel heights, and that is all that is established.
In some languages, the highest vowels are particularly high (Matumbi, Sotho-Tswana and Lomwe are examples of that type). In some other languages (Kuria, Sukuma, Kikuyu) the highest vowels sound like typical IPA [i u]. This is a detail that is generally glossed over, especially in IPA-compliant transcriptions. The earliest transcriptional practices (esp. Meinhof, Guthrie, and usually Meeussen) for proto-Bantu recognised this special property of the highest vowels by writing the highest vowels of the proto-language as [i̧,u̧] (cedillas) or [î,û]. This is a transcriptional reification of a spirantizing tendency of these vowels, that the highest vowels often cause a preceding vowel to somewhat spirantize (e.g. *ti̧→tsi, si; ku̧→fu), whereas the 2nd degree vowels don't do this.
The third degree vowels are generally pronounced closer to [ɛ ɔ] in all languages (5 and 7 vowel languages). The choice of symbols for the second-highest vowels is most variable, and in fact you can find competing practices in the same language, where some scholars transcribe the degree-2 vowels as [ɪ ʊ] and some select [e o]. You will find that in closely-related languages such as Kikuyu versus Kikamba, the degree-2 vowels are similar but noticeably and systematically distinct (in Kikuyu they more closely matches IPA [e o] and in Kamba they are between [e o] and [ɪ ʊ]). Within Logoori, some speakers use a more [e,o]-like pronunciation and some use a more [ɪ,ʊ]-like pronunciation.
It is actually important to inquire into the native language and linguistic training of the analyst in this matter, because anglophones favor [ɪ ʊ] and francophones favor [e o]. In other words, the phonetics of the degree 2 vowels is sufficiently problematic that we can't even decide, in a given language, which IPA letter is most appropriate. Therefore, only trust the numbers, and don't trust the letters. (Unfortunately, people usually don't report any numbers).
As far as Bantu-internal evidence is concerned, I don't think there are any particularly compelling arguments pointing to one of these phonetic interpretations over the other. I should point out that Swahili is a 5 vowel language, and it does not have [ɪ ʊ]. There is a general pattern in Bantu that the highest and second-highest vowels do, synchronically, become glides before another vowel, so both /CiV, CɪV/ → [CyV], regardless of the phonetic realization of the second-degree vowels. However: the whole story is a bit more complicated, because the full 7 vowels only contrast in the root-initial syllable, and otherwise /ɛ ɔ/ don't exist in prefixes or suffixes, except via harmony. If we would the highest two degrees "acting together" and excluding the lower degree, that would classically be evidence that those vowels have something in common, for example, they are [+high] (implying /i ɪ ɛ/). The same pattern is also consistent with /i e ɛ/: the highest vowels are [+ATR]. In the case of glide formation, not only can the target class be specified via a single feature no matter what the presumed vowel system, but there is no actual need to exclude any vowels from undergoing the rule.
This article may assist you an an introductory level: it has a section on vowel harmony, and a pointer to Hyman 1999 which is, I would say, the best historical account of Bantu height harmony. This dissertation by Leitch presents the Congo Basin situation (covering Lingala inter alia). Some standard references on Bantu historical linguistics are Meinhof Grundzüge einer vergleichenden Grammatik der Bantusprachen, Guthrie Comparative Bantu, Meeussen "Bantu grammatical reconstructions".
As for the connection to Niger-Congo, it is generally accepted that Volta-Congo had 10 surface vowels and there is some suggestion that Atlantic-Congo might have as well (based on evidence for 10 vowel systems in Atlantic); see Williamson in The Niger-Congo Languages. She points out that the vowels [ə e o] may be purely derived, via ATR harmony – a situation that exists in some Bantu languages as well (such as Kinande). The best work on reconstructions is done at the lower levels, such as Benue-Congo, and I think claims about Niger-Congo are premature (I would not say that the 10 vowel theory is widely accepted, rather, it is widely recognised as a possibility). Benue Congo, an ancestor of Bantu, almost certainly had a 7 vowel system.
