Review remediation: floor Short ID allocation + doctor detection

[andreiRS]

From the three-lens deep review (review-remediation). Theme: floor Short ID allocation. Conforms to ADR-0006.

Problem

ADR-0006 says Short IDs are "monotonically increasing... never reused." createTask (src/store.ts:88-102) reads next_id from meta.yaml verbatim with no floor against the real max. Any drift (bad merge, partial restore, manual edit) yields a duplicate-id file; findTask returns the first match, so show/mv/set/rm/link silently operate on the wrong task. doctor detects none of it.

Slices

6. Short ID allocation floored against the real maximum (finding #4 — allocation)

Allocate next_id = max(metaNextId, maxExistingId(live ∪ archived) + 1) with a pre-write uniqueness check.

• With meta.yaml's next_id seeded below the real max, tasks new never creates a duplicate Short ID.
• A normal sequence of tasks new still yields monotonic, never-reused ids.
• A new that would collide picks the floored id instead.

7. doctor reports duplicate Short IDs (finding #4 — detection)

Teach tasks doctor to detect and report tasks sharing a Short ID.

• With two files sharing a Short ID, doctor flags the duplication and names the colliding id(s).
• A healthy store reports no duplicates.

Open question

Flooring next_id scans live ∪ archived on every create — confirm the cost is negligible at expected human/agent task volumes.

Method

TDD outside-in at the CLI boundary. One commit per green. Both slices can start immediately.