ADR-097: --describe describes a command's output in every mode

--describe answers what a command returns, offline and whatever other flags are given, in the description member (ADR-096). For a command it holds the JSON Schema of the whole document for a run and for a dry run, each as outputSchema, with what --dry-run does for the command and at which tier. For a group or the root it lists the commands and what --dry-run does for each, so one call covers the whole tool. A path that names no command gets its error inside description and exits 0 under --json or --yaml; in text it is the usage error it looks like, exit 2. In text, --describe prints for people an outline of the fields a run writes, with their types, allowed values and descriptions, marking those that can be absent, and says what --dry-run does. The help line for --describe says it describes the output.

Derive the schemas, the outline and the dry-run behaviour from the command's result type and its dry-run classification (ADR-010); never write them by hand. Render every outline from its schema with one renderer, internal/cli/outline. Do not add arguments or flags to the description: --help covers input.

A caller needs the shape of the output before it runs anything, and that is the part --help does not give. A schema of data alone cannot validate a document, and says nothing about what --dry-run returns. The outline serves a person, and an agent that only needs the shape, at a fraction of the schema's length.

Not chosen

• Arguments and flags in the description: --help already gives them, and each argument would need a description written by hand.
• The schema as bare JSON when neither --json nor --yaml is given: JSON without --json, and not the envelope either; a person gets a wall of schema and a validator a wrapper.
• A new flag name such as --output-schema: A deprecation cycle for a flag explained wherever it appears; the key and the help line say output.