fix(mass): convergence_func on PowerLawBroken, PowerLawMultipole, cNFW family

[Jammy2211]

Summary

Follow-up to #466, which added convergence_func to the dPIE family. A planning-time audit flagged ~12 more profiles; on inspection most already inherit a working convergence_func (DevVaucouleurs, Exponential, GaussianGradient, SersicGradient, the NFW MCR variants). Three families genuinely lacked it and are reachable via radial mass integration (mass_integral → mass_angular_within_circle_from → Einstein radius) and/or MGE three_D=False potential decomposition, where the abstract convergence_func raised NotImplementedError:

• PowerLawBroken(+Sph) — extracted a _convergence radial helper shared by convergence_2d_from and a new convergence_func.
• PowerLawMultipole — convergence_func returns the zero monopole (a pure multipole encloses zero net azimuthally-averaged mass).
• cNFW family (cNFW, cNFWSph, cNFWMCRScatterLudlow(Sph), cNFWMCRLudlow(Sph)) — convergence_func reuses the MGE-of-3D-density decomposition radially; q-independent like NFW.convergence_func.

Separately, PowerLawBroken.potential_2d_from now raises a clear NotImplementedError. Its MGE potential decomposition integrates the convergence along a complex contour, which requires an analytic convergence; the broken power law's convergence is piecewise (kink at break_radius) so the result was numerically invalid (~1e7). It already raised NotImplementedError before this PR (via the missing convergence_func); this makes the failure explicit and documented. Convergence, deflections, convergence_func, and the Einstein-radius / enclosed-mass integrals remain correct.

API Changes

Added convergence_func overrides on PowerLawBroken, PowerLawBrokenSph (inherited), PowerLawMultipole, and the cNFW family (cNFWSph + MCR variants inherit). PowerLawBroken._convergence is a new private radial helper. PowerLawBroken.potential_2d_from now raises NotImplementedError explicitly (it was already non-functional). No signatures of existing public methods changed. See full details below.

Test Plan

• python -m pytest test_autogalaxy/profiles/mass/ — 413 passed
• New regression tests in test_power_law_broken.py, test_power_law_multipole.py, test_cnfw.py (convergence_func matches radial helper / zero monopole / q-independence; mass integral runs; potential raises)
• End-to-end via autolens_workspace_test/scripts/mass/{total,dark}.py: convergence_func enclosed-mass cross-check (mass_angular = πR·α) PASS for PowerLawBrokenSph (rel 1.7e-11), cNFWSph (rel 1.5e-4), PowerLawMultipole (0); cNFW/cNFWSph self-consistency all PASS

Full API Changes (for automation & release notes)

Added

• PowerLawBroken.convergence_func(grid_radius, xp=np) — radial κ(r); inherited by PowerLawBrokenSph
• PowerLawBroken._convergence(radii, xp=np) — private radial helper shared with convergence_2d_from
• PowerLawMultipole.convergence_func(grid_radius, xp=np) — returns zero (zero monopole of a pure multipole)
• cNFW.convergence_func(grid_radius, xp=np) — MGE-of-3D-density radial convergence; inherited by cNFWSph, cNFWMCRScatterLudlow(Sph), cNFWMCRLudlow(Sph)

Changed Behaviour

• PowerLawBroken.potential_2d_from / PowerLawBrokenSph now raise NotImplementedError with an explanatory message (the MGE potential is numerically invalid for the profile's piecewise convergence). Convergence, deflections, and Einstein-radius integrals are unaffected.

🤖 Generated with Claude Code

[Jammy2211]

Workspace PR: PyAutoLabs/autolens_workspace_test#136