feat: simulate_substructure end-to-end jittable simulator

[Jammy2211]

Summary

• Adds simulate_substructure to autolens/lens/substructure_util.py — a pure jnp function that chains scan-based multi-plane ray-tracing (prompt 2), source light evaluation, FFT PSF convolution, and Poisson noise via jax.random.PRNGKey
• Supports prng_key=None to skip noise for deterministic comparisons
• The complete theta -> noisy_image pipeline compiles under jax.jit

API

from autolens.lens.substructure_util import simulate_substructure

image = simulate_substructure(
    grid, image_shape, halo_params, halo_mask, scaling_matrix,
    macro_deflections_fn, macro_plane_mask, sheet_kappas,
    source_image_fn, psf_kernel, exposure_time, background_sky_level,
    prng_key=jax.random.PRNGKey(42),
    halo_profile_cls=ag.mp.NFWTruncatedSph,
)

Test plan

• Pre-convolution lensed image matches existing Tracer path (atol=1e-4)
• jax.jit(simulate_substructure) compiles and matches eager result
• Poisson noise produces different, non-NaN image vs clean
• Different PRNGKeys produce different noise realizations
• Companion PR: autolens_workspace_test#128

Ref: PyAutoLens#542 (prompt 3 of 4)

🤖 Generated with Claude Code