fix: eval_grad_tree_array works with SubArray inputs

[MilesCranmerBot]

Fixes a crash when calling eval_grad_tree_array with X::SubArray (e.g. view(dataset.X, :, idx) used by batching/custom losses).

Empirical reproducer (on master):

• eval_grad_tree_array(tree, view(X, :, idx), options; variable=true) throws a TypeError because we typeassert the gradient to typeof(X).
• DynamicExpressions returns an Adjoint{T, Matrix{T}} gradient for SubArray inputs.

Change:

• Relax the expected gradient type in InterfaceDynamicExpressions.eval_grad_tree_array to AbstractMatrix{eltype(X)}.
• Add a regression test that exercises eval_grad_tree_array on view(X, :, idx) and compares against Zygote.

Related: #362.

[github-actions]

Benchmark Results (Julia v1)

Time benchmarks

	master	1b3a01c...	master / 1b3a01c...
search/multithreading	14.7 ± 0.16 s	15 ± 0.26 s	0.985 ± 0.02
search/serial	32 ± 0.049 s	31.8 ± 0.18 s	1.01 ± 0.006
utils/best_of_sample	1.74 ± 0.42 μs	1.6 ± 0.4 μs	1.09 ± 0.38
utils/check_constraints_x10	17 ± 4.4 μs	16.9 ± 4.3 μs	1 ± 0.36
utils/compute_complexity_x10/Float64	2.18 ± 0.093 μs	2.15 ± 0.082 μs	1.01 ± 0.058
utils/compute_complexity_x10/Int64	2.08 ± 0.09 μs	2.05 ± 0.09 μs	1.02 ± 0.062
utils/compute_complexity_x10/nothing	1.52 ± 0.08 μs	1.52 ± 0.08 μs	1 ± 0.074
utils/insert_random_op_x10	5.39 ± 1.7 μs	5.4 ± 1.7 μs	0.998 ± 0.45
utils/next_generation_x100	0.435 ± 0.025 ms	0.447 ± 0.025 ms	0.973 ± 0.077
utils/optimize_constants_x10	0.034 ± 0.0075 s	0.0336 ± 0.0073 s	1.01 ± 0.31
utils/randomly_rotate_tree_x10	8.33 ± 0.96 μs	8.42 ± 0.96 μs	0.989 ± 0.16
time_to_load	2.59 ± 0.0098 s	2.63 ± 0.039 s	0.982 ± 0.015

Memory benchmarks

	master	1b3a01c...	master / 1b3a01c...
search/multithreading	0.204 G allocs: 52.5 GB	0.205 G allocs: 53.4 GB	0.983
search/serial	0.207 G allocs: 53.8 GB	0.207 G allocs: 53.8 GB	1
utils/best_of_sample	0.038 k allocs: 3.25 kB	0.038 k allocs: 3.25 kB	1
utils/check_constraints_x10	0.034 k allocs: 0.875 kB	0.034 k allocs: 0.875 kB	1
utils/compute_complexity_x10/Float64	0 allocs: 0 B	0 allocs: 0 B
utils/compute_complexity_x10/Int64	0 allocs: 0 B	0 allocs: 0 B
utils/compute_complexity_x10/nothing	0 allocs: 0 B	0 allocs: 0 B
utils/insert_random_op_x10	0.046 k allocs: 1.94 kB	0.045 k allocs: 1.88 kB	1.03
utils/next_generation_x100	4.63 k allocs: 0.276 MB	4.63 k allocs: 0.276 MB	1
utils/optimize_constants_x10	24.5 k allocs: 25.4 MB	21.5 k allocs: 21.3 MB	1.19
utils/randomly_rotate_tree_x10	0.042 k allocs: 1.34 kB	0.042 k allocs: 1.34 kB	1
time_to_load	0.145 k allocs: 11 kB	0.145 k allocs: 11 kB	1

[MilesCranmerBot]

Resumed this branch and revalidated it against current master.

What changed:

• keep eval_grad_tree_array return type concrete by normalizing the gradient to Matrix{eltype(X)}
• preserve the existing regression test for SubArray batched inputs

Local verification:

• minimal batched-view reproducer now returns Matrix{Float32} gradient and succeeds
• TEST_GROUP=integration/ad/zygote Pkg.test(coverage=true) passes
• full local Pkg.test() passes (1464 tests)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!