Mangled reified GADT record selector

From @RyanGlScott (originally #16980 (comment 222424), but unrelated to that ticket):

Consider this program:

{-# LANGUAGE GADTs #-}
{-# LANGUAGE TemplateHaskell #-}
module Main where

import Language.Haskell.TH

$([d| data T b where
        MkT :: { unT :: a } -> T a
    |])

$(return [])

main :: IO ()
main = putStrLn $(reify 'unT >>= stringE . pprint)

If you run this, you'll get a surprising answer:

$ /opt/ghc/8.8.1/bin/runghc Bug.hs
Main.unT :: forall (a_0 :: *) . Main.T b_1 -> b_1

The reported type is forall a. T b -> b, which is completely bogus! Even stranger is that if I load this module into GHCi and then reify it:

$ /opt/ghc/8.8.1/bin/ghci Bug.hs -XTemplateHaskell
GHCi, version 8.8.1: https://www.haskell.org/ghc/  :? for help
Loaded GHCi configuration from /home/rgscott/.ghci
[1 of 1] Compiling Main             ( Bug.hs, interpreted )
Ok, one module loaded.
λ> import Language.Haskell.TH
λ> putStrLn $(reify 'unT >>= stringE . pprint)
Main.unT :: forall (a_0 :: *) . Main.T a_0 -> a_0

Then the reported type is forall a. T a -> a, as expected. I can't help but wonder if this is another case of metavariables leaking through when they shouldn't.