Fixes #74.

What

Six workspace commands were sync #[tauri::command] fns calling populate_git_info, which shells out to 5–8 git subprocesses. In Tauri v2 a sync command runs on the GTK main thread, so each call froze the entire UI (~100–300 ms warm cache, unbounded if a git subprocess wedges).

• create_workspace, create_empty_workspace, get_or_create_home_workspace, create_workspace_with_preset, refresh_workspace_git_info, checkout_default_branch_in_workspace → async fn, git work via the existing populate_git_info_async (blocking pool).
• checkout_default_branch (slow on large repos) additionally wrapped in spawn_blocking.
• create_workspace_impl becomes async; its control-socket caller (dispatch_request) was already async and now awaits it.
• No State/non-Send guard held across an .await (only owned PathBufs cross the spawn_blocking boundary — same shape as commands/git.rs and create_worktree_workspace_impl).
• Frontend invoke() already returns a Promise → zero caller changes.
• package-lock.json: version field synced to 0.8.0 (stale since the bump).

Verification

• cargo check ✅
• cargo test ✅ — 1677 passed; the 2 failures (project_codemux_entry_is_filtered_out, resolve_binary_finds_native_binary_from_project_root) fail identically on the unmodified baseline (machine-local MCP config / sandbox), pre-existing.
• npm run check ✅, npm run test ✅ (1818 tests)

E2E (real app, real IPC, pathological git)

Booted the actual built app in an isolated environment (own XDG_RUNTIME_DIR/config/data/HOME, dedicated identifier + devUrl) with a git shim on PATH that sleeps 1.5 s per invocation (simulating a huge repo / cold cache), then drove create_workspace through the control socket against a real repo while probing concurrently:

• create_workspace took 22.6 s of wall-clock git work (shim exercised) and succeeded; the new workspace snapshot carried real git info (branch=main, changed_files=1).
• 59 samples of the GTK main thread's kernel wait channel while git children were in flight: every sample was S / poll_schedule_timeout (healthy event-loop park). Never wait4/pipe_read — the signature a sync handler would show while reaping a git child.
• Concurrent IPC probes during the 22.6 s run: p50 = 1 ms, p90 = 1 ms — runtime fully responsive throughout.
• Webview booted and hydrated against the patched backend (real frontend over real IPC).