P1: Add TOON encoder for JSON-array outputs (gh, kubectl, aws) #3

Context

Token-Oriented Object Notation (TOON, https://github.com/toon-format/toon) is a schema-aware JSON replacement designed for LLM prompts. It cuts ~40% of tokens on arrays of uniform objects by emitting the schema once in a header (users[3]{id,name,role}:) and then rows.

Many gh/kubectl/aws/gcloud/az outputs are exactly this shape (--json, -o json, --output json, --format json). Squeez currently runs them through CloudHandler which only strips noise — never reformats. Free 40% on a large class of common dev commands.

Goal

Add an opt-in TOON re-encoding step to the CloudHandler (and any other JSON-emitting handlers later). When the captured output is a JSON array of flat, uniformly-shaped objects, emit TOON in place; otherwise pass through unchanged.

Acceptance criteria

• New module src/strategies/toon.rs with:
  • try_to_toon(text: &str) -> Option<String> — zero-dep parser + emitter.
  • Returns Some(toon) only when input is a top-level JSON array of flat objects with identical key sets in identical order, and all values are primitives (string / number / bool / null).
  • Returns None for any deviation — never produces lossy output.
• CloudHandler calls it before its existing pipeline; if Some, returns the TOON lines.
• Unit tests cover: gh PR list shape, kubectl pods, aws ec2 describe, heterogeneous keys (must reject), nested values (must reject), single object (must reject), empty array.
• Zero new runtime deps.

Out of scope

• Streaming JSON (NDJSON from docker ps --format '{{json .}}') — handled in a follow-up.
• cargo metadata — nested structures, not worth TOON.