pylint silently skipped generative_processes: missing __init__.py removed a whole subpackage from scope

[ealt]

Surfaced by #198.

What happened

simplexity/generative_processes/ had no __init__.py — it was an implicit namespace package.
As a result pylint simplexity (the CI command) never walked into it: the whole subpackage, 21
modules and ~3,750 lines, was silently outside pylint's scope. Confirmed by comparison:

# at main (4075750)
$ pylint simplexity tests                                            # 10.00/10, 0 messages
$ grep -c factored_generative_process <output>                        # 0  <- never analyzed
$ pylint simplexity/generative_processes/factored_generative_process.py   # 3 messages

#198 adds __init__.py to carry the pending-deprecation warning, which brought the subpackage into
scope and surfaced 11 pre-existing messages:

Code	Count	Files
too-many-locals (R0914)	5	factored_generative_process.py, structures/{fully_conditional,conditional_transitions,sequential_conditional}.py
too-many-arguments (R0913)	2	factored_generative_process.py, independent_factored_generative_process.py
unused-argument (W0613)	2	structures/{fully_conditional,conditional_transitions}.py (both the context parameter, required by the ConditionalStructure protocol)
too-many-instance-attributes (R0902)	1	factored_generative_process.py
invalid-name (C0103)	1	independent_factored_generative_process.py (T_i, notation-faithful)

#198 suppresses these via per-file-ignores rather than fixing them, because the package is pending
deprecation and refactoring code slated for removal is poor value. That is a deliberate deferral, not
a claim they are wrong.

Why it is worth a ticket rather than just leaving the ignore

Two separate things, and the second is the more interesting one:

1. The messages themselves. Mostly benign for numerical code — too-many-locals in a scan over
   per-factor tensors is not really a smell, and T_i is deliberate notation. The two
   unused-argument hits are protocol-required parameters, which is a legitimate case for a narrow
   inline disable. If this package survives (see Decide the home of mixed-state presentation before deprecating generative_processes #194 / Decide whether transition_matrices.py is deprecated alongside generative_processes #195 on which parts do), the ignore should be
   narrowed to what is genuinely intended and the rest fixed.

2. The scoping gap is the real bug. A missing __init__.py silently removed a large subpackage
   from linting, and nothing surfaced that — the score stayed a clean 10.00/10, which reads as "fully
   linted" when it was not. Worth checking whether anything else is invisible for the same reason:

   $ find simplexity -type d -not -path '*/__pycache__*' -exec test ! -e {}/__init__.py \; -print
   

   A CI check that pylint's analysed-module count matches the file count would prevent a recurrence.
   This is the detection gap, and it is the part that generalises beyond this package.

Suggested resolution

• Audit for other directories missing __init__.py and fix them, expecting more surfaced messages.
• Add a guard so linter scope silently shrinking is visible.
• Once Decide the home of mixed-state presentation before deprecating generative_processes #194 and Decide whether transition_matrices.py is deprecated alongside generative_processes #195 settle which parts of generative_processes survive, narrow or remove the
  per-file-ignores entry accordingly — the surviving code should be lint-clean on its merits, and
  the deleted code makes the question moot.