Burn down the example-notebook parameter-coverage backlog (~94 methods / 257 params)

[breimanntools]

Problem

Example notebooks must demonstrate every public parameter of the method they document
(.claude/rules/notebooks.md). This was never enforced, so a large backlog accumulated: when
the new CI gate tests/unit/api_tests/test_notebook_param_coverage.py was added, it found
~94 methods with ~257 undemonstrated public parameters across the non-prediction notebooks.
Those are currently recorded in tests/unit/api_tests/notebook_param_coverage_baseline.txt so
the gate is shippable (prediction notebooks were fixed to zero gaps; the rest are baselined).

A baselined param is a documented, public parameter that its example notebook never shows — a
silent hole in the executable docs.

Goal

Burn the baseline down to empty: every public parameter of every method with an example
notebook is demonstrated (as an explicit keyword, grouped sensibly), and the baseline file is
deleted.

Scope / requirements

• Work through notebook_param_coverage_baseline.txt subpackage by subpackage (each row is
  Symbol::method::param). For each: add the param to its notebook as an explicit kwarg in a
  "further parameters" cell, re-execute with saved outputs (aa.display_df for tables,
  plt.tight_layout() + plt.show() for plots), and remove the corresponding baseline
  row(s) in the same change (the staleness guard enforces this).
• The gate may only shrink the baseline, never grow it. New notebooks/params must ship with
  zero gaps (already enforced).
• A handful of genuinely non-demonstrable params (visual-only passthroughs) may move to the
  ALLOWLIST in the test with a one-line reason instead of the baseline — keep this tiny.

Non-goals

• No change to the gate's mechanism (already in place).
• No behavioural code changes — this is notebook + docs work only.

Measurable acceptance (KPIs)

• notebook_param_coverage_baseline.txt reaches 0 rows and is removed; ALLOWLIST holds
  only visual-only passthroughs, each with a reason.
• test_notebook_param_coverage.py passes with the baseline empty.
• Every touched notebook executes cleanly under pytest --nbmake with saved outputs.
• Suggested increments: one PR per subpackage (e.g. sf_*, stp_*, sm_*, plot_*, …) so
  each is reviewable, each dropping its rows from the baseline.

Context

Follow-up to the AAPredPlot eval / notebook-coverage PR, which fixed the 14 prediction
notebooks and introduced the gate + baseline. The recurring root cause was that the notebook
param-coverage checker existed only as a skill script, never wired into CI; it is now a real
gate, and this issue tracks paying down the debt it surfaced.