feat(console): drill into a k8s fleet's roster instead of declining

[brettchien]

Summary

studio#146 slice 4 of 4 — the final slice, and the console-side fix for the originally-reported bug: clicking a k8s-runtime fleet in the Fleets panel now actually opens Fleet detail and loads its roster, instead of selectFleet showing an info toast and refusing to switch.

• source.ts: listDeployments/runtimeContext gain an optional fleet param, forwarded to the Tauri bridge (feat(src-tauri): forward a fleet arg through the roster/identity bridge commands #150) alongside cluster.
• main.ts: selectFleet branches on fleet.runtime instead of declining for k8s — activeCluster becomes an unused "" sentinel for a k8s fleet (reads now go by fleet name, which oab-mcp resolves to the bound context/namespace, feat(studio-cp,oab-mcp): k8s roster observe + runtime_context dispatch #149), and the header label shows context/namespace instead of a cluster string. tick() now always passes activeFleet as fleet to listDeployments — a no-op for the no-fleet-selected case, and for an ecs fleet this only makes oab-mcp's existing member-side filtering redundant with the console's own filterByMembers (same result, not a behavior change).
• types.ts: RuntimeContext.cluster is string | null now (null for a k8s-runtime fleet, with context/namespace set instead); FleetBinding gained optional context/namespace fields alongside the existing profile/region. runtimeContext() has no console call site yet (true before this change too) — this is type-contract parity with oab-mcp's tool, not new UI surface.

Not in scope (matches the issue body): start/stop (scale) for a k8s roster row isn't wired — scale() still calls scaleDeployment without a fleet arg, so it'll hit the ecs-only deploy_scale path and fail with a clear error toast (not silently) if attempted against a k8s deployment. That's a separate action-surface follow-up, not blocking this read-only roster view.

Test plan

• npm test — 107/107 passing (no new tests: this is control-flow wiring over already-tested pure functions — filterByMembers/serviceName already have k8s-shaped-data coverage from the delete-fleet/member-name-preview work)
• npm run typecheck — clean
• npm run build — clean
• Manual: no live k8s cluster in this sandbox to exercise end-to-end — same caveat feat(studio-cp,oab-mcp): k8s roster observe + runtime_context dispatch #149 already carries

Closes #146.

🤖 Generated with Claude Code