Separate recurring task instances into their own store

[janitorr]

Implements the separate-recurring-task-instances OpenSpec change.

Summary

• Add a dedicated recurring_task_instances store/table and remove RecurringTaskId from one-shot tasks.
• Make the recurring template the aggregate root: every /tasks/recurring/{id}/... endpoint resolves {id} as the template id. complete marks the template's current active instance done and creates the next.
• Strip recurrence logic from the one-shot complete handler so it depends only on ITaskStore and IClock.
• Morning report: add a type discriminator (one-shot | recurring) to each item, report recurring under the template id, and bump schemaVersion to "3".
• Update MCP complete_recurring_task to take the template id from list_recurring_tasks and return the completed recurring instance.

Verification

• dotnet build Nagger.slnx succeeds.
• dotnet test Nagger.slnx: 134 tests pass (74 Core + 60 Host).
• Core mutation testing: 79.65% (above the 75% break threshold); newly-introduced mutants are all killed.

Note: this change builds on add-seven-day-report-window, which is already archived on main.

[janitorr]

Reviewed. No logic bugs found in the core change; the design and implementation are coherent with the OpenSpec proposal. A few actionable issues:

1. USAGE.md is now stale (Medium — docs)
AGENTS.md points to USAGE.md as the canonical API reference, but the breaking semantics aren't reflected there:

• USAGE.md:174 still says POST /tasks/recurring/{id}/complete takes "the instance id" (now the template id).
• USAGE.md:179 still says "Completing a non-recurring task ... returns 400" — that path no longer exists.
• USAGE.md:185 still says to use /tasks/one-shot "to discover the generated instance ids" — recurring instances are no longer in one_shot_tasks.
• USAGE.md:205 still shows "schemaVersion": "2" and item objects without the new required type field.

2. Canonical specs not synced (Medium — consistency)
openspec/specs/recurring-task-creation/spec.md and openspec/specs/recurring-task-lifecycle/spec.md still describe the old behavior ("creates a one-shot task instance", "complete ... for a recurring task instance", "Reject completion of a non-recurring task"). The deltas exist under openspec/changes/..., but the canonical specs now disagree with the implementation. Expected pre-archive for OpenSpec, but confirm whether /opsx-sync should run here or after merge.

3. Migration drops RecurringTaskId with no data relocation (Medium — data loss risk, acknowledged non-goal)
20260817213759_SeparateRecurringTaskInstances.cs does DropColumn on one_shot_tasks and creates the new table without moving rows. Any existing nagger.db will have its recurring instances orphaned as plain one-shot tasks (still active) and its templates left with zero instances, so complete returns the "no active instance" validation error. The design explicitly declares "no production data exists" as a non-goal, which is fine, but SQLite is the canonical local store and persists between runs — worth confirming no reviewer/dev has data that would silently degrade.

4. stryker-config.json change is unrelated to this feature (Low — scope)
The reporters change (Html → Markdown/Json) is a separate commit (43cc01a) bundled into this branch. It has nothing to do with the instance-store change and should ideally land as its own PR.

5. IRecurringTaskInstanceStore.GetByIdAsync is unused (Low — nit)
Declared in Ports.cs and implemented in SqliteRecurringTaskInstanceStore, but no handler calls it (complete/pause/resume/cancel all use GetByTemplateIdAsync). The design lists it, so this may be intentional for future instance-by-id endpoints; if not, it's dead code.

Core handlers, status mapping, morning-report classification, and exception-to-404 mapping all look correct, and the added tests cover the breaking cases (complete-by-template, no-active-instance, missing-template).

[janitorr]

Thanks for the thorough review. Addressing each point:

1. USAGE.md stale — fixed. Updated the recurring-task section (complete now takes the template id and returns the recurring instance; removed the non-recurring/instance-id references; list no longer points at /tasks/one-shot) and the morning-report example (schemaVersion "3" and the new required type field on each item, including a recurring example).

2. Canonical specs not synced — fixed. Ran the OpenSpec sync workflow: recurring-task-creation, recurring-task-lifecycle, one-shot-task-lifecycle, and morning-task-report main specs now reflect the new behavior. openspec validate --all passes (12/12). The deltas still live in the change; the canonical specs now agree with the implementation.

3. Migration drops RecurringTaskId with no data relocation — intentional. This is an explicit non-goal in design.md (No data migration of existing rows (no production data exists)). Agreed it's worth flagging: anyone with an existing nagger.db that created recurring tasks before this change would lose those instances on upgrade. No change here per the design.

4. stryker-config.json commit unrelated — intentional. The reporters change (Html → Markdown/Json) was made deliberately to make Stryker output machine-parseable for future mutation-testing runs; it's bundled on this branch on purpose.

5. GetByIdAsync unused — intentional. The design specifies the IRecurringTaskInstanceStore port with GetByIdAsync for future instance-by-id endpoints; all current handlers use GetByTemplateIdAsync.