Slim down lineark usage for LLM token efficiency

[flipbit03]

Summary

• Collapse low-frequency commands (teams CUD, labels CUD, comments manage, relations, documents CUD, milestones, embeds) into single-line entries with (--help for flags) hints
• Keep high-frequency commands fully expanded: issues CRUD/search/list, comments create, all read/list commands
• ~110 lines → ~75 lines (~30% token savings on every LLM invocation)
• Zero loss in discoverability — everything is one --help away

Before/After example

# Before (12 lines for labels)
lineark labels create <NAME>                     Create a label
  [--team KEY] [--color HEX]                     Team, color
  [--description TEXT]                           Description
  [--parent-label-group ID]                      Nest under a group label
  [--make-label-group]                           Create as a group label
lineark labels update <ID>                       Update a label
  [--name TEXT] [--color HEX]                    Name, color
  ...5 more lines...
lineark labels delete <ID>                       Delete a label

# After (1 line)
lineark labels create|update|delete ...          Manage labels (--help for flags)

Test plan

• make check passes (fmt, lint, clippy, docs)
• make test passes (103 offline + 55 unit tests)
• Column alignment lint passes

Closes #122