Add primitive_wall reference, tilted-gravity packing strategy, servo …

[cgoniva]

…kp tuning

New references/commands/primitive_wall.md: when to prefer primitive_wall over mesh, and a pointer to its own docs for restrictions (no motion mechanism) and output behavior (no VTK output) -- both since documented directly in primitive_wall.html's own Restrictions section.

STRATEGIES.md: three new entries -- verifying a per-particle-state command actually took effect (mark_particles/define_group/enable_heat_transfer can run with no error while doing nothing), tilting gravity during a non-physical prep phase to pack particles at a target location for free, and mesh_module servo's kp often needing to be much larger than its default to actually reach maximum_velocity.

RULES.md: one-line addition cross-referencing the verification entry above under Simulation Output.

variable.md: merged the existing boolean-only ${name}-nesting warning with a newly confirmed equal-formula case (same underlying bug, extends to equal/atom/boolean formulas alike) into one section instead of two.

[d-louw]

Pushed a follow-up round of fixes on top of the original commit:

• primitive_wall.md: rewrote it from a one-line stub into a real reference (what it's for, what it can't do — no motion mechanism, shear is surface-velocity only — shape/keyword overview, restart/output notes) and linked it from SKILL.md's Guidelines, per CLAUDE.md. Added an Examples section (verbatim shape examples from the docs, plus sheared-floor and rotating-drum shear examples). Also softened the "no VTK output" claim to a flag-for-verification since it isn't actually in the docs' Restrictions section (which just says "none").
• STRATEGIES.md: moved the "verify a per-particle-state command" entry below the file's own intro paragraph (it had landed above it). Pulled the insertion-related entries (until_filled/until_settled mode choice, verifying a pack target, the rate_in_region recipe) plus the tilt-gravity and mesh_module servo kp entries into a new references/strategies/INSERTION.md, since the tilt-gravity and servo entries turned out to be notes from the same compacted-bed workflow and the rate_in_region recipe is a full walkthrough — same pattern as the existing MESH_PREPROCESSING.md.
• RULES.md: the new dump-ordering line pointed at a strategy entry about a different failure mode (verifying a command took effect vs. ordering relative to the last output write). Reworded so the rule stands on its own (using write_output_timestep instead of the legacy "dump" term) and reframed the cross-reference to explain why they're related.
• variable.md: the merged nesting-rule section implied equal/atom/boolean formula style was the deciding factor for when ${name} fails; it's actually whether the formula is a quoted string, confirmed against:

  variable y equal 1-${x}     # valid
  variable y equal "1-${x}"   # invalid
  variable y equal 1-v_x      # valid, dynamic
  variable y equal "1-v_x"    # valid, dynamic
  

  Restored the "known Aspherix bug" framing and the concrete boolean example that the merge had dropped, and added the quoted/unquoted example to make the actual rule explicit.
• Reflowed all new prose to one sentence per line per CLAUDE.md's formatting rule, and fixed -- → — to match the surrounding style.