BUG: Matfree EquationBC with Schur fails

[KarsKnook]

When using EquationBC in a matrix-free context with Schur complement preconditioning, we do not get the correct behaviour (potentially the EquationBC is not enforced correctly?). For example, the problem in the MFE below should converge in 1 FGMRES iteration but it takes 2 FGMRES iterations. When commenting out "mat_type": "matfree", we get the expected behaviour.

from firedrake import *

mesh = UnitSquareMesh(2, 2)
nml = FacetNormal(mesh)

V = FunctionSpace(mesh, "RT", 1)
W = FunctionSpace(mesh, "DG", 0)
Z = V * W

z = Function(Z)
sigma, u = split(z)
tau, v = TestFunctions(Z)

F = inner(sigma, tau) * dx
F -= inner(u, div(tau)) * dx
F += inner(div(sigma), v) * dx
F += inner(div(sigma), div(tau)) * dx

bcs = [EquationBC(inner(dot(sigma - Constant([1,1]), nml), dot(tau, nml)) * ds(1) == 0, z, [1], V=Z.sub(0))]

sp = {
        "mat_type": "matfree",
        "snes_monitor": None,
        "ksp_monitor": None,
        "ksp_type": "fgmres",
        "pc_type": "fieldsplit",
        "pc_fieldsplit_type": "schur",
        "pc_fieldsplit_schur_fact_type": "full",
        "fieldsplit_0": {
            "ksp_type": "preonly",
            "pc_type": "python",
            "pc_python_type": "firedrake.AssembledPC",
            "assembled": {
                "pc_type": "lu",
                "pc_factor_mat_solver_type": "mumps",
                },
            },
        "fieldsplit_1": {
            "ksp_type": "gmres",
            "ksp_monitor": None,
            "ksp_atol": 0,
            "ksp_rtol": 1e-12,
            "pc_type": "python",
            "pc_python_type": "firedrake.MassInvPC",
            "aux_pc_type": "python",
            "aux_pc_python_type": "firedrake.AssembledPC",
            "aux_assembled": {
                "pc_type": "lu",
                "pc_factor_mat_solver_type": "mumps",
                },
            },
        }

solve(F==0, z, bcs=bcs, solver_parameters=sp)