feat(kt-config): validate_all_graph_types startup canary + API wiring

[charlie83Gs]

Summary

Adds `PluginRegistry.validate_all_graph_types()` — walks every registered graph type's composition and returns `{graph_type_id: [CompositionIssue]}`, with clean types excluded so the map is directly truthy-testable as "is the registry valid?".

Wires it into `services/api/src/kt_api/main.py::_load_plugins()` so startup logs one WARNING per broken graph type before the first pipeline task runs. Tolerant, not fatal: a missing contribution is an operator-visible config bug (expected during phased plugin extraction while provider ids still point at inline implementations), not a reason to refuse to boot.

Log shape: one line per broken graph type, every unresolved id on that line, so ops can grep one string per plugin instead of spelunking through dozens of log entries.

Delegates per-type to the existing `validate_composition` so pair-wise agentic + concept_extractor + list semantics stay consistent in one place.

Test plan

• 4 new tests in `test_validate_composition.py`: empty registry, clean registry returns `{}`, broken types grouped by id, mixed registry excludes clean types
• `libs/kt-config`: 212 pass
• `services/api`: 80 unit pass (startup path exercised by existing app import)
• `ruff check`/`format` clean, pyright green
• pre-commit + pre-push lefthook green

🤖 Generated with Claude Code

[charlie83Gs]

Addressed review feedback

Widened to cover the main gap — canary now fires from every process that dispatches pipeline tasks, not just the API.

#1 Canary wired to workers 🔴 — Extracted the three-step bootstrap triad (load → bridge → validate+log) into a single `bootstrap_plugins()` function in `kt_providers.registry`. Replaced the identical copy-paste at 5 entry points:

• `services/api/src/kt_api/main.py`
• `services/worker-search/src/kt_worker_search/main.py`
• `services/worker-bottomup/src/kt_worker_bottomup/main.py`
• `services/worker-ingest/src/kt_worker_ingest/main.py`
• `services/worker-all/src/kt_worker_all/main.py`

Every process that might dispatch a pipeline task now fires the canary at boot. Side benefit: de-dupes the 5-way copy-paste the review flagged.

#2 `reason` preserved 🟡 — Kept `{i.reason}` verbatim in the log message, as you noted. Added a test assertion (`test_bootstrap_logs_canary_for_broken_graph_type`) that explicitly checks the reason string surfaces in the WARNING so this doesn't drift.

Nit: drop quotes on return annotation — Done. `PluginRegistry.validate_all_graph_types` now annotates `dict[str, list[CompositionIssue]]` unquoted; consistent with the body on L1071.

Nit: tests reuse helper — Considered, left as-is. The new `test_bootstrap_plugins.py` is a separate file using lightweight stubs; reusing `_FullPlugin` from `test_validate_composition.py` would create a cross-file dependency for minor LoC savings. Inline stubs keep the two test files self-contained.

Tests

• 3 new tests in `libs/kt-providers/tests/test_bootstrap_plugins.py`: bridging end-to-end, canary logs for broken graph type (with reason preserved), empty registry is silent
• `just test-unit` across all 24 packages: green (including the 4 worker unit test suites that newly import `bootstrap_plugins`)

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}