Add richer Keeper review-status tracking beyond reviewed/not-reviewed

Title
Add richer per-row status tracking for ACP-based Keeper review artifacts

Body

Problem

Resume currently works by detecting saved review files and skipping previously reviewed row_index values. That is useful, but the status model is still minimal.

Current behavior

Rows are effectively treated as:

• reviewed
• not reviewed

There is no richer per-row state for things like:

• pending
• skipped
• needs follow-up
• superseded after concept-set changes

Risk

• weak resume semantics
• hard to manage iterative review sessions
• limited auditability when concept sets or generated rows change between runs

Requested change

Introduce richer per-row review status tracking in Keeper workflow state.

Acceptance criteria

• per-row states are persisted explicitly
• resume behavior uses structured status, not only file presence
• workflow can distinguish skipped/incomplete/finalized rows
• state changes are stable across reruns where possible
• tests cover resume and mixed-status scenarios

---

7. Improve incidence-shell guidance for denominator coherence beyond TAR alone

Title
Improve incidence-shell denominator guidance across TAR, cohort entry logic, and strata settings

Body

Problem

The new incidence TAR step allows user-controlled TAR and strata settings, but denominator coherence depends on more than TAR alone.

Current behavior

The shell now provides a time_at_risk_configuration step and dialogue context, but the user guidance is still lightweight relative to the underlying analytic nuance.

Risk

Users may over-focus on TAR settings while missing denominator-relevant factors such as:

• target cohort entry logic
• first-entry behavior
• strata design
• analysis intent across comparisons

Requested change

Expand the incidence-shell guidance and dialogue context for denominator coherence.

Acceptance criteria

• the TAR step surfaces denominator-relevant context beyond TAR fields alone
• /ohdsi context includes enough metadata to discuss coherent denominator design
• prompts/help text explain when TAR is not the only driver
• tests cover presence of the new context/help wiring

If you want, I can turn these into a tighter “ready to paste into GitHub” format with labels, priority, and suggested issue ordering.

Start by locating the Keeper workflow state and resume logic, then inspect the incidence-shell time_at_risk_configuration step and its /ohdsi dialogue context. Review existing tests before changing either area; done means explicit per-row statuses drive stable resume behavior, and denominator guidance/context is covered for mixed-status and help-wiring scenarios.