Removing DHE is a mistake. The discrete log in prime fields is fine---as fine as RSA is, anyway---and it's a handy PFS backup in the (unlikely) case deployed elliptic curves turn out to be significantly wounded.
Your first sentence seems to me like an excellent reason to remove DHE altogether from TLS 1.3, considering those servers do not support, and presumably may never support, (the draft) Finite Field DHE parameter negotiation.
Discrete log in prime fields does have the index calculus problem; it won't keep being good forever, and the performance gets worse. I'm banking on having enough different backup between ECDHE over secp256r1 and X25519 over Curve25519 that any elliptic curve difficulty won't be a problem.
Index calculus. Over prime fields it has seen essentially no major progress (beyond small complexity tweaks, some of which are useful) since 1992 with the number field sieve. Index calculus also exists for elliptic curves, under some conditions: once again, over prime fields things seem fine (modulo MOV, anomalous, etc curves). I suspect we will also have to drop RSA if the index calculus for prime field discrete logs ever improves significantly. Likewise, some efficient attack against P-256 or curve25519 has a good chance to eliminate most or all curves in that size range.