DevWebUI 0.6.0

Added

• Folders and .devwebui files can be opened directly. The new devwebui open <path>
  command is the drag-and-drop entry point used by the Windows launcher. A dropped folder is
  scanned for configured or detectable projects; new projects are registered, while dropping an
  already-registered project starts it.
• Runtime selection now follows each project's lockfile. With the global runtime set to
  auto, Bun lockfiles select Bun and npm, Yarn, or pnpm lockfiles select Node. An explicit
  per-process runtime remains the highest-priority override.
• Project configuration reloads live. Changes to a registered project's .devwebui file are
  detected and applied without requiring a daemon restart.

Changed

• Plain Bun, Node, and executable commands start without a permanent shell wrapper. Commands
  that need shell syntax or a .cmd/.bat shim still use the shell, while safely tokenizable
  executable commands launch directly. This removes the extra cmd.exe process previously shown
  for every compatible managed server on Windows.
• Machine scanning is quieter after setup. A new installation performs one discovery scan so
  the first dashboard is useful, then leaves automatic startup scans disabled unless the user
  explicitly enables them.
• High-volume logs use bounded batching. Live log rings, SSE delivery, and disk persistence
  now trim and flush per batch instead of repeatedly shifting arrays or performing a synchronous
  append for every child-output event.

Fixed

• The live error feed matches the current-process view. Errors from an earlier daemon session
  or an older run no longer reappear through SSE after the snapshot and HTTP views have filtered
  them out.
• Updater integration tests use the actual runtime executable. This keeps JavaScript arguments
  intact on Windows installations where bun on PATH is an npm command shim.
• Project watching no longer depends on fs.watch supplying a filename. Events from platforms
  that omit it still trigger the configuration re-check.
• Daemon restarts keep a healthy instance available. Instance probing retries transient
  failures, and tray health checks require consecutive failures before treating the daemon as
  dead.