feat: TM1 chore as a first-class task kind (closes #156)

[nicolasbisurgi]

Summary

Implements TM1 chore as a first-class task kind alongside TI process (closes #156). A task now carries exactly one of process or chore — the field name is the kind discriminator (no meta-kind field). Mixed process + chore taskfiles are supported across JSON, TXT, and cube sources, and chores participate in the DAG (predecessors, stages, retry) on equal footing with processes.

Anchor docs (in-repo working artefacts; only CONTEXT.md is committed):

• CONTEXT.md — Task / Task kind / Chore vocabulary (committed in this PR).
• docs/adr/0002-polymorphic-task-kinds.md — full decision record (local-only).
• IMPLEMENTATION_PLAN_CHORE_KIND.md — file-by-file change list (local-only).

What's in scope

• Polymorphic TaskDefinition / Task / OptimizedTask with mutual-exclusion as a class invariant on Task.__init__ and per-kind forbidden-field validation in validate_task.
• Chore-specific validation: tm1.chores.exists(...) for every unique (instance, chore_name), plus a SINGLE_COMMIT check when safe_retry: true (deduplicated so the chore object is fetched at most once per run).
• execute_task dispatch to _execute_chore_task + new execute_chore_with_retries. Binary success/fail (HTTP 204 / exception). safe_retry re-executes the whole chore.
• TXT taskfiles now run through validate_taskfile on read (closes a pre-existing gap).
• Cube reader accepts chore rows; rejects rows with both process and chore set.
• Additive cube schema upgrade: chore measure element + vchore TI variable + CellPutS line in both v11 and v12 }rushti.load.results bodies.
• Additive SQLite schema: chore TEXT column on task_results, PRAGMA-driven migration on first connection.
• Optimizer calculate_task_signature is chore-aware; signatures occupy a disjoint hash space from process signatures.
• Dashboard + DAG visualization render a unified Task target column with [P] / [C] kind indicator (both visualization_template.py and templates/visualization.html updated to stay consistent).
• Structured execution logger gains a parallel chore: Optional[str] parameter (does not overload process).
• CHANGELOG documents the two behavioural changes.

Out of scope

• Shell command task kind (a separate Issue B).
• Other ideas from Run other commands #145 (sub-taskfiles, REST, reusable stages).

Test plan

• Unit tests in tests/unit/test_chore_kind.py (37 cases) cover validation, the Task invariant, TXT conversion + validation tightening, chore signature disjointness, retry behaviour with mocked TM1Service, execute_task dispatch, validate_tasks chore branch, and the cube reader's chore acceptance + both-populated rejection.
• tests/unit/test_taskfile.py updated to match the new "exactly one of 'process' or 'chore'" message.
• Full unit suite passes: 772 passed, 5 skipped.
• Integration tests against live TM1 v11.8 (tm1srv01) in tests/integration/test_v11_chore_kind.py — 13/13 pass. Uses preconfigured test_chore_success and test_chore_error chores. Verifies:
  • execute_chore_with_retries: success (single attempt), failure without safe_retry (single attempt), failure with safe_retry (exhausts budget then raises).
  • validate_tasks chore branch: existence pass/fail, SingleCommit check on safe_retry.
  • execute_task dispatch routes chore-kind tasks to the chore path on success and on failure.
  • Mixed process + chore DAG with predecessors crossing kinds — forward path executes in order, failing chore aborts dependent tasks via require_predecessor_success.
  • build_logging_objects idempotently adds the chore measure element to an existing rushti cube.
  • Stats DB records process and chore rows with disjoint signatures (no collision when they share a name).
  • End-to-end results push: chore execution row → stats DB → upload_results_to_tm1 → }rushti.load.results TI execution → cube cell readback under the chore measure. This proves the v11 TI body compiles against the new vchore variable, the CSV column ordering aligns with PROCESS_VARIABLES, and the additive cube schema works in a single round-trip.
• v12 integration coverage not exercised here — same code paths but worth re-running against tm1srv02 before merge.

Upgrade notes

• Re-run rushti build --tm1-instance X on existing installations to add the new chore measure to the cube. The additive merge logic already in place handles this without data loss.
• Flush pending CSVs from prior versions before upgrade — the }rushti.load.results TI variable list changed and old-format CSVs cannot be loaded.
• Pre-existing malformed TXT files will now fail with explicit validation errors on first run. This is net-positive (the files were silently buggy before) but worth flagging to users with legacy taskfiles.

🤖 Generated with Claude Code