skillscript-runtime v0.32.0

Upgrade impact: none (additive). Dashboard rendering only; no runtime or API change.

Sharpens the "What it does, step by step" diagram so a non-programmer approver can see a skill's blast radius at a glance — the acceptance bar being "could they, from the diagram alone, correctly say what it touches and what it changes?" (if not, approval is rubber-stamping).

• Risk-first weighting. Steps are now rendered by effect tier so the eye lands on what matters: mutations (data_write / file_write / skill_write / notify) get a colored highlight band + bold; shell (runs a command) gets a band + bold; external effects (data/file reads, model calls, connector dispatch, execute_skill) sit at normal weight; plumbing ($set / emit / json_parse — internal var-shuffling with no external effect) is recessed (muted + smaller). Previously every step had equal weight, so the one mutation could hide among a stack of Set X steps.
• Plain-language branch conditions. Conditions are humanized instead of showing raw skillscript: ${UNITS} == "metric" → "If UNITS is metric", ${RAW|contains:"current_condition"} → "If RAW contains current_condition", not …|contains:"github.com" → "… does not contain github.com".
• Truncated rows are recoverable. A row cut with … now carries a hover tooltip with the full text (and the → VAR it produces), so the approval-relevant tail isn't lost.