Add JSON-string interop helpers: encode_json() and loads()

[jeepinbird]

Summary

Implements the ergonomic helpers from upstream issue #49 / PR #57, adapted to this slimmed codebase (not a verbatim port).

Data frequently arrives as raw JSON text — LLM tool outputs, REST API responses, logs — where the JSON null keyword has no direct TOON equivalent. Worth noting: this is an ergonomic win rather than a correctness fix. Our encode() already renders Python None as the TOON null literal, and json.loads already maps null → None. These helpers just remove the boilerplate of wiring the two together for the common JSON-string → TOON path.

What's added

• loads(json_string, **kwargs) — a thin wrapper over json.loads returning TOON-ready Python objects (null → None). Extra kwargs are forwarded.
• encode_json(json_string, options=None) — one-step JSON string → TOON, equivalent to encode(loads(json_string), options).

from toon_format import encode_json

encode_json('{"name": "Alice", "mood": null, "tags": [1, null, 3]}')
# name: Alice
# mood: null
# tags[3]: 1,null,3

Implementation notes

• Both helpers live in a new json_io.py module that imports encode() directly. This sidesteps the circular-import problem the upstream PR ran into (it placed the helpers in the same module that defines encode, requiring careful ordering). Since we removed utils.py during the slim-down, a dedicated module is the natural home.
• Exported from the public API (toon_format.__init__).
• New tests in tests/test_json_integration.py cover nulls in objects/arrays, top-level null, kwargs forwarding, option forwarding, equivalence to the manual encode(json.loads(...)) pipeline, and JSONDecodeError on invalid input.
• README and docs/api.md / docs/README.md updated.

Verification

• Full suite: 805 passed, 13 skipped (818 collected; skips are by-design normalization fixtures).
• ruff check src/ tests/ clean.

---

Generated by Claude Code