Inconsistent validity checking for unnamed vs. named wildcards

GHC HEAD rejects the following program:

{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE PartialTypeSignatures #-}
{-# LANGUAGE NamedWildCards #-}
{-# LANGUAGE PolyKinds #-}
{-# OPTIONS_GHC -Wno-partial-type-signatures #-}
module Bug where

import Data.Proxy

class C k (a :: k)

f :: (C _ a) => Proxy a
f = Proxy

GHCi, version 8.9.20190220: https://www.haskell.org/ghc/  :? for help
Loaded GHCi configuration from /home/rgscott/.ghci
[1 of 1] Compiling Bug              ( Bug.hs, interpreted )

Bug.hs:12:9: error:
    Wildcard ‘_’ not allowed in a constraint
      except as the last top-level constraint of a type signature
        e.g  f :: (Eq a, _) => blah
      in the type signature for ‘f’
   |
12 | f :: (C _ a) => Proxy a
   |         ^

But it accepts it if you change _ to _k, a named wildcard!

f :: (C _k a) => Proxy a
f = Proxy

This feels strangely inconsistent to me, as I would have expected GHC's treatment of these two programs to be the same.