Clear code-quality warnings; make application-view e2e assertions real

[nabramovitz]

Clears all 46 open warning-severity findings on the Code Quality board (the level that holds both dimensions at "Fair"), fixing the real bugs behind them rather than suppressing.

Backend (30 findings)

Every Go warning was a genuinely dropped error:

• initSessionStore dereferenced the store returned by all three session-store constructors before checking err — a constructor failure was a nil-deref panic at startup.
• The helm job restWorker read response.StatusCode before the err check. DoProxySingleRequest always returns a nil response with a non-nil error, and the worker runs in its own goroutine, so this was a whole-process crash. Errored jobs now carry an Err marker; UpdatePods skips its stale-pod sweep and UpdateResources keeps the cached resource when a job failed, so a transient proxy error no longer empties the release view.
• The kube terminal ignored the getClients error (nil-client panic at secretClient.Create) and assigned its pod-wait timeout error to a loop-shadowed variable, reporting a never-started pod as success.
• Three kubeconfig auth providers (azure, kubeconfig, oidc) dropped the ParseKubeConfig error, so a malformed kubeconfig upload panicked jetstream; awsiam dropped the getTokenIAM error and persisted an empty token as success.
• unarchive dropped the write-side Close error (lost-data class); the cfapppush no-op init method is deleted; the dashboard tryDecode* error discards are now explicit.

Frontend (16 findings)

• SetPage's constructor assigned the keepPages parameter instead of the parameter property, so forceLocalPage never actually forced keepPages on the dispatched action.
• Smaller items: uninitialized-then-always-overwritten local in the firehose formatter, two missing spaces in concatenated user-facing strings, a duplicated property in the proxy config template, and a collapsed null/undefined comparison.

E2E: real assertions instead of hollow ones

Ten application-view tests asserted expect(boolExists !== undefined) — always true for a boolean, so they verified nothing. Each now either asserts real behavior or skips with the reason it can't be tested here:

• Services unbind / details: seed a user-provided service instance and binding through the CF API, then assert the bound instance's card and its Unbind row action. New createUserProvidedService / deleteServiceInstance helpers (labeled stratos-e2e-test and covered by the cleanup sweep, which now reaps service instances before spaces).
• Variables edit / delete: seed a variable, assert the Edit/Delete row actions behind the signal-list kebab.
• Events filtering: assert the signal-list name-filter input.
• SSH / per-instance logs: skipped — they need a running instance and the API-created test app has no bits.
• Log auto-scroll / level filter / download: skipped — the modern log viewer auto-scrolls by default and has no filter or download UI (these two features were not carried over from the Material UI; worth separate issues if they should return).

Giving these tests teeth exposed two latent harness bugs the hollow assertions had been hiding: updateAppEnvironment used POST where CF API v3 requires PATCH, and openRowActionMenuByRow returned a menu bound to the removed Material menu selector instead of the row's own actions menu. bindService now returns the binding guid (guarded for managed bindings, which return 202 with an empty body).

Verification

• make lint: golangci-lint 0 issues on both Go modules; make check gate green (eslint, 3094 frontend tests, production build, go tests).
• The five rewritten e2e tests pass and five skip at 2 workers against a live local stack; the remaining application-view failures predate this change (stale pre-modernization selectors, tracked separately).

Shared-URL banner correctness (found during live verification)

Live-verifying this branch against a dev instance with 4 CF + 4 Kubernetes shared-URL endpoints surfaced a mislabelled banner: the home page fed every connected endpoint into the duplicate-URL counter and reported "8 Cloud Foundry endpoints share a URL" when only 4 CF endpoints exist. Duplicate detection now runs per endpoint type, and both banners (home and the Application Wall / Marketplace / Services one) build their sentence from a single shared helper so the phrasing cannot drift: "4 Cloud Foundry endpoints share a URL." / "4 Cloud Foundry endpoints and 4 Kubernetes endpoints share URLs." The wall banner's page-specific explanation moves to its own line. Included here rather than as a follow-up so operators picking up this fix level get the banner correction with it.

Endpoint management kebab on the CF picker + Reconnect

Also from live verification: the /cloud-foundry picker is a projection of the Endpoints page onto its CF subset, but its cards had no management menu. The per-endpoint kebab moves into a shared EndpointRowActionsService used by both surfaces (the picker omits Unregister — registration changes are admin-level and stay on the Endpoints page), and gains a Reconnect action for connected endpoints: re-auth in place through the connect dialog, prefilled with the live connection's username. Two connect-dialog fixes make that flow safe — the success detector no longer treats a pre-existing connection as an instant success (gated on hasAttemptedConnect, matching the error path), and the username prefill prefers the live connection's user over the localStorage history.

[norman-abramovitz]

LGTM