Bugfix/backsolve adjoint with componentvector

[simenhu]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

The backsolve_adjoint functionality did not work when I used a ComponentVector as a state vector since axes(::ComponentVector) returns (1:1:M, ) and not (Base.OneTo(M), ) as other array types does afaik.

Switched the axis(...) call with size(...) to get an integer instead of a range, which seems to do the trick.

MWE to reproduce:

using OrdinaryDiffEq
using ComponentArrays
using SciMLSensitivity
using Enzyme

# Mass-damper system
function mass_damper!(du, u, p, t)
    mass, spring_k, damping_c = p
    position, velocity = u.state[1], u.state[2] # BUG nr. 1
    # position, velocity = u[1], u[2]  # BUG nr. 2
    du[1] = velocity  # dx/dt = v
    du[2] = (-spring_k * position - damping_c * velocity) / mass  # dv/dt = a

    nothing
end

# System parameters
mass = 1.0      # kg
spring_k = 10.0 # N/m (spring constant)
damping_c = 2.0 # Ns/m (damping coefficient)
p = [mass, spring_k, damping_c]


u0 = ComponentArray(state = [1.0, 0.0]) # BUG nr. 1
# u0 = [1.0, 0.0]  # BUG nr. 2

# Time span
tspan = (0.0, 5.0)
dt = 0.01

# Create the ODE problem
prob = ODEProblem(mass_damper!, u0, tspan, p)

# Solve with ORK256
println("Solving mass-damper system with ORK256...")
sol = solve(prob, ORK256(), dt=dt, dense=false)

function g(u,p, t)
    # Dummy function to test the adjoint sensitivity
    return sum(u .^ 2) ./ 2
end

res = adjoint_sensitivities(sol, ORK256(); g=g, dt=dt, sensealg=BacksolveAdjoint(autojacvec=EnzymeVJP()))
@info "Adjoint sensitivity results: $(res)"