feat: Add OpenCLIDescriber interface to expose a command's OpenCLI metadata

[boblail]

Summary

Commands discovered via the OpenCLIContract parse a rich --help-opencli document (arguments, options, descriptions, examples, exit codes, hidden/interactive flags, ...), but until now that metadata was discarded — the Command interface only exposed name/summary/aliases/defaultCommand/subcommands. Consumers (e.g. the sq entrypoint) could therefore only reproduce a stripped-down OpenCLI document for commands they discover.

This adds an optional companion interface, OpenCLIDescriber, that surfaces each node's OpenCLI metadata, following the same pattern as ContractReporter (#57): the Command interface is unchanged, and consumers type-assert a discovered command to reach the richer data.

if d, ok := cmd.(exoskeleton.OpenCLIDescriber); ok {
    if c, _ := d.OpenCLICommand(); c != nil {
        // c.Arguments, c.Options, c.Description, c.Examples, ...
    }
}

Details

• commandDescriptor gains an unexported openCLI *opencli.Command field, populated by opencliToDescriptor for every node in the tree. Its Commands field is stripped — children remain reachable via Subcommands(), keeping a single source of truth for the tree.
• executableCommand threads that metadata through discover()/toCommands() and implements OpenCLICommand(). It returns nil (not a synthesized fallback) for commands discovered via any other contract.
• OpenCLICommand() returns a copy describing only the node; its Commands field is not populated.
• No change to the Command interface. Built-in commands do not implement OpenCLIDescriber.

Test plan

• go test ./...
• gofmt -l . / goimports -l . / go vet ./... clean
• New tests: TestOpenCLICommandExposesMetadata (root + hidden subcommand carrying options/arguments/hidden), TestOpenCLICommandForNonOpenCLIContract (returns nil)
• End-to-end verified against the sq entrypoint: sq <cmd> --help-opencli now reproduces the full OpenCLI document emitted by the underlying binary