Generated tabular output

[wfchandler]

Add tabular output with --format flag

Tabular output is much easier to quickly scan for value, as well as
being more amenable to traditional UNIX scripting. For example,
compare finding the block_size for an image in the following output:

  $ oxide image list --format=json
  [
    {
      "block_size": 512,
      "description": "Debian 13 generic cloud image",
      "id": "3672f476-51ff-49ab-bd2c-723151a921c6",
      "name": "debian-13",
      "os": "Debian",
      "size": 3221225472,
      "time_created": "2025-08-11T16:48:17.827861Z",
      "time_modified": "2025-08-11T16:52:23.411189Z",
      "version": "13"
    }, {
      "block_size": 4096,
      "description": "Silo image for Packer acceptance testing.",
      "id": "f13b140e-4d34-4060-b898-6316cdcc2f1e",
      "name": "packer-acc-test-silo-image",
      "os": "",
      "size": 1073741824,
      "time_created": "2025-05-15T23:11:14.015948Z",
      "time_modified": "2025-05-15T23:12:02.098119Z",
      "version": ""
    }
  ]

Versus:

  $ oxide image list --format=table:name,block_size
   NAME                        BLOCK_SIZE
   debian-13                   512
   packer-acc-test-silo-image  4096

For response types that are amenable to tabular formatting, we add a
--format flag to the subcommand. This takes an optional
comma-separated list of field names to print, as larger response items
can easily overflow typical terminal widths. The available field names
are listed in the --help output for the subcommand.

Not all API return values can be reasonably formatted as a table.
Endpoints returning (), byte streams, and unstructured
serde_json::Value objects are deliberately excluded.

Internally, this is implemented using the newly added ResponseFields
trait, which xtask creates as part of the generated CLI. This enables
getting the field names of a type and accessing them as a
serde_json::Value via their name.

[wfchandler]

Dependent on oxidecomputer/progenitor#1195

[wfchandler]

How this exposes the available fields to users.

Long help shows fields for the subcommand:

$ ./target/debug/oxide image list --help
List images which are global or scoped to the specified project. The images are returned sorted by creation date, with the most recent images appearing first.

Usage: oxide image list [OPTIONS]

Options:
      --format <format>
          Format in which to print output
          
          Possible values:
            - json                    Output as pretty-printed JSON
            - table                   Output as table with all columns displayed
            - table:field1,field2,... Output as table, specifying which columns to display
          
          Examples:
            --format json
            --format table
            --format table:name,id,description
          
          Available fields:
            - block_size
            - description
            - digest
            - id
            - name
            - os
            - project_id
            - size
            - time_created
            - time_modified
            - version

      --limit <limit>
          Maximum number of items returned by a single call

      --project <project>
          Name or ID of the project

      --sort-by <sort-by>
          [possible values: name_ascending, name_descending, id_ascending]

  -h, --help
          Print help (see a summary with '-h')

Global Options:
      --profile <PROFILE>
          Configuration profile to use for commands

Failing to provide any valid field names will show the field list:

$ ./target/debug/oxide image list --format=table:whatever
WARNING: 'whatever' is not a valid field
ERROR: None of the requested '--format' fields are present in this command's output

Available fields:
 block_size
 description
 digest
 id
 name
 os
 project_id
 size
 time_created
 time_modified
 version

[wfchandler]

I go back and forth on whether this is a good idea. As the comment says, I think having --format for write actions will just add noise to the help test.

However, this list of words to filter on is inevitably going to fall out of date, so we'll end up with most write operations not having the flag.

[wfchandler]

Clippy complains about deref for types::SwitchLinkState.

[wfchandler]

Not borrowing the field would require a clone for non-copy types.