feat: the deferred skill tranche — chain, custom analysis, simulate, plot, debug

[Jammy2211]

Closes #8 (the tranche deferred with approval from the birth build, #1). Five skills in house style, grounded in autofit_workspace scripts and PyAutoFit source:

• af_chain_searches — the three hand-off currencies (result.instance fixed / result.model posterior-updated priors / model_centred_* explicit widths) + Initializer start points, with the anti-lock-in design rule.
• af_custom_analysis — extra __init__ inputs, custom Visualizer hooks (live fit figures), custom Result.
• af_simulate_dataset — recorded-truth simulation and the recovery check that gates first real-data runs.
• af_plot_fit — the aplt roster + matplotlib-over-instances residual figures; documents that af.ex.plot_profile_1d requires a pathlib.Path (str raises TypeError on its / join — found by executing the recipe).
• af_debug_fit_failure — strict-order triage (likelihood → priors → data → sampler) with the af.Drawer uninformative-data cross-check.

Gates (supervised --auto, launch 2026-07-10; ship sign-off human, in-session)

• Suite 53/53 · symbol audit 0 misses · 96 citation paths resolve · idiom lint clean · provenance --strict 0/0.
• Every code recipe executed against the installed stack.
• Heart YELLOW reason set human-acked at launch (recorded in Mind active.md).

The assistant's skill set is now the complete planned roster: 19 skills.

🤖 Generated with Claude Code