Typechecker plugins aren't run when doing pattern-match checks

Very common code with type-level natural numbers:

{-# OPTIONS_GHC -fplugin GHC.TypeLits.Normalise #-}

data Vec :: Nat -> Type -> Type where
  Nil  :: Vec 0 a
  (:>) :: a -> Vec n a -> Vec (n + 1) a

infixr 5 :>

merge :: Vec n a -> Vec n a -> Vec (n + n) a
merge Nil       Nil       = Nil
merge (x :> xs) (y :> ys) = x :> y :> merge xs ys

(or zip, zipWith etc).

This will, with -Wincomplete-patterns, lead to spurious "incomplete pattern match" errors.

What happens during pattern match checking is that we try to find whether the following Givens are inconsistent:

[G] n1 ~ 0
[G] n2 + 1 ~ n1

If we ran the usual type-checker plugins on this, we'd expect these Givens to be reported as inconsistent.

However, this wasn't happening. After a quick investigation, it seems that the tcg_tc_plugin_solvers field of TcGblEnv is empty in this context. Here's what I see in the tc-trace:

solveSimpleGivens {
  [[G] pm_d10l_d10l {0} :: n1 ~# 0 (CNonCanonical)]
Emitting fresh work
  {[G] pm_d10l_d10l {0} :: n1 ~# 0 (CNonCanonical)}
----------------------------- 
Start solver pipeline {
  tclevel = 0
  work item = [G] pm_d10l_d10l {0} :: n1 ~# 0 (CNonCanonical)
  inerts = {Type-function equalities = {[G] co_a10c {0}
                                              :: n2 + 1 ~# n1}
            Innermost given equalities = 0
            Given eqs at this level = True}
  rest of worklist = WL {}
[...]
End solver pipeline }
runTcPluginsGiven: no solvers -- trace I added in runTcPluginsGiven
End solveSimpleGivens }

So it looks to me that typechecker plugins are incorrectly initialised for calls within the pattern match checker.