docs(skill): bias brainstorm toward minimal solutions and deletion cost

[codeaholicguy]

Summary

• add a Baseline step: state the simplest honest solution in 1–3 sentences before diverging; every option must justify what it adds over it, and the baseline stays a candidate
• add two mandatory Challenge axes: speculative generality (every abstraction/layer/flag/data copy needs a current caller) and verified claims (load-bearing "needed" claims must be checked against runtime behavior, not doc assertions)
• add deletion cost to Compare: on ties prefer the smaller option — complexity gets added on demand but is rarely removed voluntarily
• add a self-review gate before Converge: ask "why do we need this?" for each new constant/object/layer/flag; cut messaging-only or future-proofing answers
• note "scope without a current caller" in the Avoid list

Rationale

Review arcs on #147 (capacity) and #196 (remove-registry) showed late-cut complexity that a baseline-first, deletion-cost-aware brainstorm would have caught at design time: a cache no runtime consumer read, multi-provider scaffolding with one real provider, a 77-line frozen registry-id snapshot used only for message wording, and a mutation object with unused fields. The lesson: always solve with the simple and minimal solution, and treat deletion cost as a first-class criterion.

Validation

• Markdown-only change; no code paths affected