MethodError on `length(::DiffResults.ImmutableDiffResult{1, Float64, Tuple{Float64}})`

Hello, here to report another possible bug found upstream in Turing. In the following example, f differentiates just fine but g doesn't.

import ReverseDiff

f(x) = exp.(x[])
f([1.0])
ReverseDiff.gradient(f, [1.0])

g(x) = exp.(reshape(vec(x), ()))
g([1.0])
ReverseDiff.gradient(g, [1.0])

Edit: h(x) = exp.(fill(x[], ())) works fine too, but j(x) = exp.(reshape(x, ())) doesn't.

Stack trace

julia> ReverseDiff.gradient(g, [1.0])
ERROR: MethodError: no method matching length(::DiffResults.ImmutableDiffResult{1, Float64, Tuple{Float64}})
The function `length` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  length(::Cmd)
   @ Base process.jl:716
  length(::Base.MethodSpecializations)
   @ Base reflection.jl:1317
  length(::Core.SimpleVector)
   @ Base essentials.jl:933
  ...

Stacktrace:
  [1] _similar_shape(itr::DiffResults.ImmutableDiffResult{1, Float64, Tuple{Float64}}, ::Base.HasLength)
    @ Base ./array.jl:652
  [2] _collect(cont::UnitRange{…}, itr::DiffResults.ImmutableDiffResult{…}, ::Base.HasEltype, isz::Base.HasLength)
    @ Base ./array.jl:711
  [3] collect(itr::DiffResults.ImmutableDiffResult{1, Float64, Tuple{Float64}})
    @ Base ./array.jl:705
  [4] broadcastable(x::DiffResults.ImmutableDiffResult{1, Float64, Tuple{Float64}})
    @ Base.Broadcast ./broadcast.jl:707
  [5] broadcasted
    @ ./broadcast.jl:1318 [inlined]
  [6] broadcast(f::ReverseDiff.ForwardOptimize{…}, x::ReverseDiff.TrackedArray{…})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/elementwise.jl:237
  [7] broadcast
    @ ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/elementwise.jl:198 [inlined]
  [8] _materialize
    @ ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/broadcast.jl:265 [inlined]
  [9] materialize
    @ ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/broadcast.jl:273 [inlined]
 [10] g(x::ReverseDiff.TrackedArray{Float64, Float64, 1, Vector{Float64}, Vector{Float64}})
    @ Main ./REPL[6]:1
 [11] ReverseDiff.GradientTape(f::typeof(g), input::Vector{…}, cfg::ReverseDiff.GradientConfig{…})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/tape.jl:199
 [12] gradient(f::Function, input::Vector{Float64}, cfg::ReverseDiff.GradientConfig{ReverseDiff.TrackedArray{…}})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/gradients.jl:22
 [13] top-level scope
    @ REPL[8]:1
Some type information was truncated. Use `show(err)` to see complete types.

Version info

(ppl) pkg> st
Status `~/ppl/Project.toml`
  [37e2e3b7] ReverseDiff v1.15.3

julia> versioninfo()
Julia Version 1.11.1
Commit 8f5b7ca12ad (2024-10-16 10:53 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 10 × Apple M1 Pro
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, apple-m1)
Threads: 1 default, 0 interactive, 1 GC (on 8 virtual cores)

Start by running the reported Julia reproducer and trace the failure through src/derivatives/elementwise.jl and src/derivatives/broadcast.jl, with the tape entry points in src/api/tape.jl and src/api/gradients.jl as context. Done means the reshape-based g or j example no longer raises the ImmutableDiffResult length error, with a regression test covering the behavior.