server: workspace=global requests return 500 'Path is not absolute: global' causing repeated 'failed to reload global' toasts

Summary

Desktop client shows repeated "无法重新加载 global / UnexpectedStatus" toasts because almost every workspace-scoped API returns HTTP 500 with Error: Path is not absolute: global (or empty string) when the workspace is global.

Environment

• opencode version: 0.0.0-next-17430 (desktop beta, channel=next)
• OS: macOS Darwin 25.5.0 (arm64)
• Terminal: xterm-256color
• Shell: /bin/zsh
• Install/channel: next (desktop beta app "OpenCode Beta")
• Active plugins: none

Reproduction

1. Start the desktop beta app (0.0.0-next-17430) and open the session/global workspace view.
2. The client periodically refreshes workspace-scoped endpoints; watch the log or the UI.
3. Right-bottom toast appears repeatedly: 「无法重新加载 global … UnexpectedStatus」.

Expected Behavior

The session list and global workspace should load normally (HTTP 200).

Actual Behavior

Session.list passes the workspace identifier (global, or empty string) into a path normalization helper that asserts absolute paths, throwing Error: Path is not absolute: global, and the endpoint returns HTTP 500. The client surfaces this as UnexpectedStatus.

Server log evidence (role=server, service process started via serve --service):

http.method=GET http.url=/api/session http.status=500
cause="Error: Path is not absolute: global
    at RN (../core/src/database/path.ts:19:15)
    ...
    at Session.list (../server/src/handlers/session.ts:56:39)"

Within a single refresh cycle the following endpoints all returned 500 (33 occurrences total): /api/agent, /api/command, /api/location, /api/mcp, /api/mcp/resource, /api/model, /api/model/default, /api/permission/request, /api/project/current, /api/provider, /api/question/request, /api/reference, /api/session, /api/vcs, /api/workspace.

Additional Context

• Triggered whenever the client requests with workspace=global (or empty); service restart does NOT fix it — a freshly started service process reproduces the same 500s on the next client refresh cycle.
• Same root cause also appears with an empty path (Path is not absolute: ), suggesting a shared path-resolution path used across handlers.
• Also observed in the same log: SQLiteError: no such column: workspace.project_id (Workspace.list) and repeated SQLiteError: disk I/O error (157 occurrences), possibly related DB schema/state inconsistencies.