refactor(data): scope async resource consumers

[beruro]

Problem

Shared application loaders could publish late results after project, workspace, session, or browser scope changes. Some polling/countdown work also continued while hidden, browser diagnostic state could outlive the owning surface, and Tauri event consumers could leak subscriptions when cleanup ran before asynchronous listener registration resolved.

Solution

• Migrate application, project, browser, git, settings, policy, skill, terminal, and sidebar consumers to the scoped async-resource lifecycle introduced by refactor(async): add scoped resource lifecycle #562.
• Fence late results by stable scope identity and suspend eligible polling/countdowns while hidden.
• Add AsyncUnlistenScope to own asynchronously-created Tauri listener handles, release late handles, roll back partial registration, and make disposal idempotent.
• Preserve the newer viewer-isolated GitHub work-items lifecycle from the base.
• Keep this PR stacked directly on refactor(async): add scoped resource lifecycle #562 and independent from sibling perf(runtime): bound caches, scans, and background work #564.

Potential risks

• This is a broad consumer sweep; an incorrect scope key can expose stale cross-project data.
• Browser close/unmount, visibility return, native menu registration, and project reactivation are the highest lifecycle-risk paths.
• Listener cleanup errors are swallowed so teardown cannot mask the originating setup error; a failing native unlisten may require platform logs.
• This PR depends on refactor(async): add scoped resource lifecycle #562 and should be retargeted only after refactor(async): add scoped resource lifecycle #562 lands. Rollback is a normal revert; there is no schema, wire, dependency, or persistence migration.

Performance audit

Verdict: pass. Each retained subscription has one Effect-owned scope; cleanup is symmetric, late handles self-release, partial setup rolls back, and no unbounded timer/cache was added.

Verification

Validated locally after merging current #562 head 8618af370 into head 49864e065 without conflicts:

• pnpm install --frozen-lockfile — PASS. The optional legacy node-sass native postinstall warning did not block installation or checks.
• Focused Vitest across the 7 changed/foundation lifecycle files — PASS, 7 files / 28 tests.
• pnpm typecheck — PASS.
• git diff --cached --check before the merge commit — PASS.

Because this stacked PR targets junyu/async-resource-lifecycle, full frontend/Rust CI may not run until #562 lands and this PR is retargeted to develop.