CLI: human-readable --pretty output rendering

Optional: add a `--pretty` flag that renders stdout as human-readable text instead of JSON. Only worth doing if a non-agent consumer materializes (dev debugging, direct human use).

Parent: #407
Status: speculative — open only if a real use case appears.

## Why not MVP

Agent is the only consumer during the hackathon. All help/error/output text is agent-read. Pretty rendering doubles the per-command output code path for marginal value.

## If needed

- Per-command renderer function next to the JSON serializer.
- Keep JSON as default; `--pretty` opts in.
- Probably only needed for `markets`, `balance`, `position`, `quotes` where tabular output is meaningfully better.

No files or tests are named. Start by locating the CLI command entry points and JSON serializers for markets, balance, position, and quotes, then assess whether a real non-agent consumer justifies the work. Done means JSON remains the default and an opt-in --pretty mode renders the relevant outputs as human-readable text.

Written by the indexing model from the issue text.