feat: add loom doctor command for config validation

[subotic]

Problem

When config values are invalid or stale (e.g., terminal.command = "ghostty" on macOS where the binary doesn't exist), loom fails at runtime with an unhelpful error. There's no way to proactively validate the config.

Proposal

Add a loom doctor command that validates the config and reports issues with suggested fixes. Checks should include:

• Binary existence: verify that terminal.command and editor.command resolve to actual executables or valid open -a patterns
• Scan roots: verify paths in registry.scan_roots exist and contain git repos
• Workspace root: verify workspace.root exists and is writable
• Agent config: validate MCP server names, plugin format (name@marketplace), marketplace repos are accessible
• Sandbox config: check that allowed paths exist
• Preset references: verify any preset names referenced in workspace manifests still exist in config

Example output

$ loom doctor
Checking config...
  [ok] workspace root /Users/foo/workspaces exists
  [ok] scan root /Users/foo/_github.com exists (found 42 repos)
  [warn] terminal.command "ghostty" not found in PATH
         hint: on macOS, use "open -a Ghostty" instead
  [ok] editor.command "zed" found in PATH
  [ok] 2 agent plugins validated
All checks passed (1 warning).

Context

Came up when an existing config had terminal.command = "ghostty" which doesn't work on macOS (needs open -a Ghostty). The PR #69 fix for detect_terminal() only helps new configs created via loom init. A doctor command would catch stale configs proactively.