feat(BA-6882): expose kernel scheduling history via repository/service/adapter/REST v2/SDK/CLI

[jopemachine]

Exposes kernel_scheduling_history end-to-end through the v2 backend + REST + SDK + CLI, mirroring the existing session/deployment/route slices. GraphQL is intentionally out of scope — BA-6883 adds it on top of the adapter/service built here.

What was already there

The table, ORM row, domain data types, and an admin-only search_kernel_history repository method already existed. Everything above the repository boundary was missing, and the repository had no scoped variant.

Layers added

Layer	Added
models/scheduling_history/conditions.py	Kernel MatchSpec filter variants + string / datetime / result conditions
models/scheduling_history/orders.py	KERNEL_ORDER_FIELD_MAP, default/tiebreaker orders, resolve_kernel_order
common/data/permission/types.py	EntityType.KERNEL_HISTORY, KERNEL_SCOPED_HISTORY
repositories/scheduling_history/	KernelSchedulingHistorySearchScope + search_kernel_scoped_history (repository + db_source)
common/dto/manager/v2/scheduling_history/	KernelHistoryFilter / Order / Node, KernelHistoryScopeDTO, admin + scoped inputs, payload
services/scheduling_history/	2 actions, 2 service methods, processor wiring
api/adapters/scheduling_history/adapter.py	admin_search_kernel_history, kernel_scoped_search, querier/filter/order/DTO conversion
api/rest/v2/scheduling_history/	POST /kernels/search, POST /kernels/scoped/search
client/v2 + client/cli/v2	search_kernel_history / kernel_scoped_search, bai scheduling-history kernel search / search-scoped

Two deliberate deviations from the existing slices

Both follow the documented forward direction, which the existing session/deployment/route code predates.

1. The scope is a body field, not a path param. A kernel query is scoped by session_id, kernel_id, or both, and an empty scope must be rejected — none of which a path param can express. api/rest/v2/AGENTS.md names this exact shape as the forward direction (/v2/{entity}/scoped/search + scope as a request body field), and client/v2/AGENTS.md names the matching scoped_search(request) SDK form. Hence the route /kernels/scoped/search and the scoped_ handler prefix.

Non-emptiness is validated once, in the shared DTO model_validator (KernelHistoryScopeDTO), and defended again in the repository scope.

2. The scoped route uses auth_required, not superadmin_required. api/rest/v2/AGENTS.md states that all scoped search routes use auth_required, and this story requires a non-admin scoped search. Note the existing session/deployment/route scoped routes are all registered superadmin_required, which appears to defeat the purpose of a scoped search — left alone as out of scope here.

The CLI follows from (1): search-scoped takes --session-id / --kernel-id rather than a positional ID, and errors cleanly when neither is given.

Notes

• sub_steps is intentionally absent from KernelHistoryNode: kernel_scheduling_history has no such column (the 84b901f69d16 migration added it to deployment/route only). BEP-1061 and BA-6852 both describe kernel as having sub_steps — that discrepancy needs resolving on the writer side.
• Overlaps fix(BA-6887): reference the real from_status/to_status columns in kernel history conditions #12857 (BA-6887) on ~14 lines of conditions.py: this branch carries the same from_phase → from_status correction, since it is the first caller of those factories. Whichever lands second needs a trivial conflict resolution.
• Live CLI verification is not yet possible. kernel_scheduling_history has no writer, so the table is empty; BA-6852 adds it. Every other acceptance criterion is met in code.

🤖 Generated with Claude Code

[jopemachine]

Superseded by a stack of smaller PRs, one per layer: #12867 (BA-6889), #12868 (BA-6890), #12869 (BA-6891), #12870 (BA-6892), #12866 (BA-6893). Same change, reviewable in ~300-450 line pieces.