fix: use analytic potential for truncated NFW

[Jammy2211]

Summary

Replace the MGE approximation in NFWTruncatedSph.potential_2d_from with the exact analytic potential for the Baltz, Marshall & Oguri (2009) n=1 truncated NFW profile.

Use a sixth-order small-radius series to avoid catastrophic cancellation in NumPy/JAX float32, retain a finite zero-centred potential convention, and correct the paper's arXiv identifier.

Fixes #563.

API Changes

NFWTruncatedSph.potential_2d_from now returns the analytic BMO potential instead of the previous MGE approximation.
A reusable dimensionless analytic helper, potential_func_sph_from, is added.
See full details below.

Test Plan

• pytest test_autogalaxy/profiles/mass/dark/test_nfw_truncated.py -q — 48 passed
• pytest test_autogalaxy/ -q — 1054 passed
• Full-mode 200×200 workspace checks for all three tNFW truncation regimes — 9 PASS / 0 FAIL / 0 SKIP
• Independent comparison against lenstronomy TNFW
• JAX JIT, autodiff, float32 stability, centre, x=1, and large-truncation checks
• GitHub Actions Tests and Docs workflows

Gate Record

PR creation was explicitly authorized despite unrelated PyAutoHeart RED reasons from another task's shared checkout:

• PyAutoFit: 2 commit(s) behind origin
• PyAutoGalaxy: on branch feature/point-source-light (not main)
• PyAutoGalaxy: 1 commit(s) behind origin

After CI passed, the human explicitly authorized merge. A fresh isolated-main Heart run removed all three RED reasons and reported STALE only because release-validation/install sidecars are unavailable in this web runtime.

Full API Changes (for automation & release notes)

Added

• autogalaxy.profiles.mass.dark.nfw_truncated.potential_func_sph_from(grid_radius, tau, xp=np) — evaluates the dimensionless analytic n=1 truncated-NFW lensing potential with stable small-radius handling.

Changed Behaviour

• autogalaxy.profiles.mass.dark.nfw_truncated.NFWTruncatedSph.potential_2d_from(grid, xp=np, **kwargs) — now uses the analytic BMO equation (18) potential rather than an MGE approximation; the paper convention sets the central additive constant to zero.

Migration

• None required; the method signature and returned physical quantity are unchanged.

Generated by the PyAutoLabs agent workflow.