UI: surface AgentPlaybookRuns in the reference client stack

[ibolton336]

Context

#36 adds the AgentPlaybook / AgentPlaybookRun controllers (the CRDs are already merged). The reference client stack from #35 (contract types, hub-shim, browser UI) is AgentRun-only today. This issue tracks the client-side work to surface playbook runs end to end.

Not blocked by #46 — this work sits entirely on the CR/status surface, which stays stable while the harness decomposition changes what runs inside stage images. Can be developed against the #36 branch on a local cluster before it merges.

Tasks

• Contract types — mirror AgentPlaybook, AgentPlaybookRun (spec + status incl. status.stages[]{name, phase, agentRunName}, currentStage) into agentrun-client types and the browser-safe agentic-client contract; add Ready-condition reason constants (PlaybookNotFound, PlaybookNotReady, StageRunning, StageSucceeded, StageFailed, StageNotFound, AgentRunCreationFailed, AgentRunDeleted, Succeeded) and label keys konveyor.io/agentplaybookrun / konveyor.io/stage. Phase reuses the AgentRun enum.
• hub-shim routes — GET /api/agentplaybooks[/:name]; GET|POST /api/agentplaybookruns, GET|DELETE /api/agentplaybookruns/:name; convenience GET /api/agentplaybookruns/:name/agentruns via the konveyor.io/agentplaybookrun label selector. On create, resolve params/models against the union of stage Agents (params forward wholesale; models apply to all stages — 400 if stages disagree on providers, since there are no per-stage overrides).
• Client methods — extend RunApi/ShimClient with playbook list/create/get/delete + stage-run listing; delete+recreate semantics only (spec is whole-spec immutable via CEL).
• Playbook-runs list view — columns per the CRD printer columns (Playbook, Phase, Current stage, Age) plus x/y stages; duration computed from startTime/completionTime (status has no duration field, unlike AgentRun).
• Detail page — stage stepper from status.stages[]; each started stage links to the existing RunDetailPage/ChatPanel for its agentRunName (stage runs are ordinary AgentRuns — ACP chat works unchanged while a stage is Running). Delete copy must state cascade: stage AgentRuns are owner-referenced and GC'd.
• Create modal — playbook picker (disable non-Ready playbooks, show AgentsNotReady message), param form from the union of stage Agents' declared params with ADR 0005 platform-resolved partitioning; no instructions field (spec has none; stage instructions come from the playbook).
• RunsPage guard — stage-owned AgentRuns (label present) appear in the flat runs list: badge them with a link to the parent playbook run, and warn/disable Delete (deleting a stage run terminally fails the parent with AgentRunDeleted, no retry).

Contract gotchas (from #36 review)

• Ready=False is the normal healthy state while a playbook run executes (StageRunning) — must not render as an error.
• Never point waitForRunning/ACP resolution at the playbook run itself (it has no sandboxName/secretKeyRef); always target status.stages[].agentRunName.
• Stage AgentRun names are <playbookrun>-<stage> but hash-truncated past 63 chars — read status.stages[].agentRunName / labels rather than recomputing.
• Cascade deletion closes any open stage chat mid-session (WS 1011) — RunDetailPage's 404 handling covers it, ChatPanel copy should too.

Out of scope (separate issues)

• Create-form model picker for plain AgentRuns (pre-existing gap, simpler post-✨ Multi-variable LLM provider credentials + sandbox pod run labels #34).
• harness-goose consuming KONVEYOR_PLAYBOOK_INSTRUCTIONS (one-line entrypoint change, needed for HITL-in-playbook).

Refs: #35, #36. Related: #22–#24 (tackle2-ui client layer inherits this shape).