fix(wcore): wait for confirmed approval mode changes

[FerroxLabs]

Changing Core from Default to an automatic approval mode currently updates Desktop’s local approval authority and reports success before Core accepts the command. A later typed refusal is dropped, leaving the permission selector and host-side approvals out of sync with Core.

This change keeps the previous mode until an accepted policy/config receipt arrives, consumes set_mode_refused, and returns the effective mode on rejection. Overlapping requests, disconnected transports and unconfirmed requests fail without widening approval authority. Temporary scheduled-run mode changes remain unpersisted. The renderer explains the refusal, refreshes main-process truth, ignores delayed results from another conversation and prevents a capability-label refresh from reinstating an old stored preference.

Related to #1223. A mode refusal does not itself disable tools; the UI says that explicitly rather than inventing a tool inventory. Producer-backed per-tool availability remains the separate #1328 contract dependency. No issue closure is requested.

Validation: exact committed tree 5e3cdf68c18faa88215a9bba3002e0ad2723f3bf passed full Hetzner verification: 21,472 Vitest tests (53 skipped), 285 Bun tests and typecheck, with no source drift. Local focused manager/decoder/renderer suites pass 76 tests. Regression tests reproduce premature local elevation and stale stored-mode restoration before their fixes; accepted and refused paths are both covered. Local typecheck, i18n validation, lint (existing 3,017 warnings, zero errors) and full prek checks pass. Remote setup errors and the initial missing nested dependency test failure were preserved before the corrected full run. Native packaged user-flow acceptance remains pending.

[FerroxLabs]

Closing as obsolete: this targets Wayland Core / Nano code that was removed in the 0.13.0 cutover. As of 0.13.1 Fuigo is the only engine, so these files no longer exist on main and the change cannot be rebased.

If the underlying behaviour still matters under Fuigo, it is worth a fresh issue against the current code rather than reviving this branch.