feat: add build orchestrator

[Blackgaurd]

Adds service/orchestrator.py with run_build() that coordinates level-by-level execution of a topological build plan.

What changed

• New service/orchestrator.py with run_build(dataset_name, version, start, end)
  • Calls schedule_build() to get a BuildPlan (Kahn's algorithm)
  • Iterates levels sequentially (barrier model)
  • Executes each job via execute_job() from the worker module
  • On failure: sets cancelled event and raises RuntimeError
  • Structured logging for level start/complete and plan summary
• New tests/service/test_orchestrator.py with 6 tests:
  • Single root works, level-by-level order correct (C -> B -> A)
  • Failure at level N prevents level N+1
  • NoValidTimestampsError propagates as RuntimeError
  • Diamond graph executes correctly (D once, {B,C} parallel-safe, A last)
  • Cancelled event is set on failure

Why

The orchestrator ties the scheduler and worker together. It owns the "execute the plan" concern, separate from "compute the plan" (scheduler) and "build one dataset" (worker).

Benefit

run_build() is the drop-in replacement for _build_recursive(). PR 6 will wire it into build_dataset() and delete the old code.

🤖 Generated with Claude Code

[Blackgaurd]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• refactor: replace _build_recursive with orchestrator #323 : 2 dependent PRs (#324 , #325 )
• feat: add build orchestrator #322 👈 (View in Graphite)
• feat: add build worker that executes a single job #321
• feat: add topological scheduler with Kahn's algorithm #320
• feat: add dependency graph collector with range expansion #319
• refactor: add job descriptor and build plan types #318
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.