Error with Turing and ReverseDiff, adding matrices

Hi -

I've been using ReverseDiff to run some models in Turing, and it appears a recent change from v1.7 to 1.8 has made my models stop working. I managed to narrow down the issue to a single line in the model, where I try to add two matrices. Oddly, it only seems to be an issue with particular combinations of matrices, but I'm not seeing the pattern, so I'm not sure what the problem is.

See MWE below:

using Turing, ReverseDiff

@model function efa(P)
    λ ~ filldist(Normal(0, 10), P, 1)
    Ψ ~ filldist(truncated(Normal(0, 5), 0, Inf), P)
    f_var ~ filldist(truncated(Normal(0, 5), 0, Inf), 1)

    Σ1 = λ * f_var * λ' + Diagonal(Ψ)        # Fails
    # Σ1 = λ * f_var * λ' + diagm(Ψ)           # Fails
    # Σ1 = randn(P, P) + Diagonal(Ψ)           # Works
    # Σ1 = λ * f_var * λ' + randn(P, P)        # Works
    # Σ1 = λ * f_var * λ' + Diagonal(ones(P))  # Works
end

Turing.setadbackend(:reversediff)
efa_test = efa(3)
@time m1 = sample(efa_test, NUTS(1, 0.65), MCMCThreads(), 1, 4)

Running this gives an error like this (only the first few lines of the stacktrace shown):

nested task error: MethodError: no method matching increment_deriv!(::Float64, ::Float64)
        Closest candidates are:
          increment_deriv!(::ReverseDiff.TrackedArray, ::Real, ::Any) at C:\Users\opus\.julia\packages\ReverseDiff\60noS\src\derivatives\propagation.jl:34
          increment_deriv!(::AbstractArray, ::Real, ::Any) at C:\Users\opus\.julia\packages\ReverseDiff\60noS\src\derivatives\propagation.jl:36    
          increment_deriv!(::AbstractArray, ::Any) at C:\Users\opus\.julia\packages\ReverseDiff\60noS\src\derivatives\propagation.jl:38
          ...
        Stacktrace:
          [1] increment_deriv!
            @ ~\.julia\packages\ReverseDiff\60noS\src\derivatives\propagation.jl:35 [inlined]
          [2] increment_deriv!(t::Matrix{Real}, x::Matrix{Float64})
            @ ReverseDiff ~\.julia\packages\ReverseDiff\60noS\src\derivatives\propagation.jl:40
          [3] special_reverse_exec!(instruction::ReverseDiff.SpecialInstruction{typeof(+), Tuple{ReverseDiff.TrackedArray{Float64, Float64, 2, Matrix{Float64}, Matrix{Float64}}, Matrix{Real}}, ReverseDiff.TrackedArray{Float64, Float64, 2, Matrix{Float64}, Matrix{Float64}}, Nothing})
            @ ReverseDiff ~\.julia\packages\ReverseDiff\60noS\src\derivatives\linalg\arithmetic.jl:50
          [4] reverse_exec!(instruction::ReverseDiff.SpecialInstruction{typeof(+), Tuple{ReverseDiff.TrackedArray{Float64, Float64, 2, Matrix{Float64}, Matrix{Float64}}, Matrix{Real}}, ReverseDiff.TrackedArray{Float64, Float64, 2, Matrix{Float64}, Matrix{Float64}}, Nothing})
            @ ReverseDiff ~\.julia\packages\ReverseDiff\60noS\src\tape.jl:93
          [5] reverse_pass!(tape::Vector{ReverseDiff.AbstractInstruction})
            @ ReverseDiff ~\.julia\packages\ReverseDiff\60noS\src\tape.jl:87
          [6] reverse_pass!
            @ ~\.julia\packages\ReverseDiff\60noS\src\api\tape.jl:36 [inlined]

Run the Julia MWE and inspect the failing matrix-addition path in ReverseDiff/src/derivatives/propagation.jl and src/derivatives/linalg/arithmetic.jl, especially increment_deriv! and special_reverse_exec!. Compare the working and failing matrix combinations, then verify that the Turing model samples successfully with ReverseDiff without the reported MethodError.