PIE Vowel System

The PIE vowel system is often refuted on the account of typology. *e - *o
opposition is not probable/possible it is said ([@] - [a] would be better
but that does not account for the comparative evidence). So only *e and *o
are considered here as "real" phonemes in a way. *e: and *o: are not, *i and
*u are just syllabic allophones of *y and *w; *ey, *oy, *ew, *ow - just
combinations of *e and *o with *y and *w. So only *e and *o are phonems
(although phonetically there are a lot more vowels) and this is of course
typologically not possible. But if we look at Sanskrit it can be analyzed
prettymuch the same way - there is only /a/, /a:/ is not regarded as a "real
phoneme" (like PIE *e: and *o:), /e/ and /o/ are diftongs (= PIE *ey, *ew
etc.), /i/ and /u/ are just syllabic allophones of /y/ and /v/. So if we
analyze Sanskrit in the same rigid way as we do with PIE we would have even
less vocalic segments than in PIE. And Sanskrit is very well recorded so we
cannot say it is typologically unfounded.
Am I missing something? Is there for instance a special reason that Sanskrit
/i/ and /u/ cannot have the same status as *i and *u in PIE? Any thoughts?

Mate