Problem

When the diff layer detects higher-level patterns like ListOfTypesDiff (and the proposed OneOfWrappingDiff from #702), the raw lower-level diffs (TypeDiff, OneOfDiff, PropertiesDiff, etc.) are still populated even though they're misleading. Every checker must then call suppression functions like shouldSuppressTypeChangedForListOfTypes() to avoid double-reporting.

This is fragile — any new checker that forgets the suppression check will produce false positives.

Proposed solution

After detecting a higher-level pattern in getSchemaDiffInternal, null out the redundant lower-level diffs at the diff layer itself:

// After computing all diffs:
result.ListOfTypesDiff = getListOfTypesDiff(value1, value2)
if !result.ListOfTypesDiff.Empty() {
    result.TypeDiff = nil
    result.OneOfDiff = nil
    result.AnyOfDiff = nil
}

result.OneOfWrappingDiff = getOneOfWrappingDiff(value1, value2)
if !result.OneOfWrappingDiff.Empty() {
    result.TypeDiff = nil
    result.PropertiesDiff = nil
    result.RequiredDiff = nil
    result.OneOfDiff = nil
}

Benefits

• The diff output accurately represents what changed (the pattern, not misleading raw deltas)
• Checkers don't need any suppression logic — the redundant diffs simply aren't there
• New checkers can't accidentally report suppressed changes
• The higher-level diff carries all the information needed

Tradeoff

Raw diffs are lost for consumers who want them, but arguably those raw diffs are wrong in these cases (e.g., TypeDiff: object → empty is misleading when it's really a oneOf wrapping).

Related: #702