Two-tier column rendering — clean threshold, flagged detail sections, severity ordering

[DEVunderdog]

Parent

#268

What to build

Extend to_markdown() with a ## Flagged Columns block containing full per-column detail sections for columns that exceed the clean threshold. A column is clean when ALL of: no MissingnessFlag, no NumericFlag, MissingSeverity is None or Minor, NonlinearityTag is None or Linear. Clean columns appear only in the Column Summary table from the previous slice — no body section. Flagged columns get a full detail section (all fields at this point — field-level trimming comes in the next slice). Flagged columns are ordered severity-first: DropCandidate/FullyNull → Severe → High → Moderate → Minor-but-flagged → numeric/nonlinearity-only flags; alphabetical within each tier.

Acceptance criteria

• Columns meeting the clean threshold produce no entry in ## Flagged Columns
• Columns exceeding the clean threshold each produce a named subsection inside ## Flagged Columns
• A column with MissingnessFlag.DropCandidate appears before a column with MissingSeverity.Severe in the output
• A column with MissingSeverity.Severe appears before one with MissingSeverity.High; High before Moderate; Moderate before Minor-but-flagged; columns flagged only for NumericFlag/NonlinearityTag appear last
• Within the same severity tier, columns appear alphabetically
• Unit tests cover: all-clean result (no Flagged Columns body), mixed result (only flagged columns get sections), ordering across all severity tiers

Blocked by

• Blocked by to_markdown() scaffold — Dataset Overview + Column Summary table #270