feat: add golden conformance test suite for format conversion and mappings

[alvinreal]

Summary

Adds a golden-file conformance test suite to detect output drift across morph's parsers, serializers, and mapping evaluator.

Tests added (14 total)

Format conversion (7 tests):

• JSON → YAML, JSON → CSV, JSON → JSON (pretty)
• CSV → JSON, CSV → YAML
• YAML → JSON, YAML → CSV

Mapping operations (4 tests):

• rename — field renaming
• select — field selection
• where — row filtering
• Complex pipeline — rename + set + drop combined

Round-trip consistency (3 tests):

• JSON → YAML → JSON
• JSON → CSV → JSON
• YAML → CSV → YAML

Structure

tests/golden/
├── inputs/
│   ├── users.json
│   ├── users.csv
│   └── users.yaml
├── expected/
│   ├── json_to_yaml.txt
│   ├── csv_to_json.txt
│   └── ... (11 golden files)
└── golden_conformance.rs

Updating golden files

After an intentional output change:

UPDATE_GOLDEN=1 cargo test --test golden_conformance

Fixes #82

[alvinreal]

CI triage: Linux/macOS are green, but both Windows jobs fail in with broad golden mismatches (11 tests).\n\nLikely cause: line-ending normalization ( vs ) in golden fixture comparisons rather than logic regressions, since round-trip tests still pass and failures are across many text outputs at once.\n\nSuggested fix:\n1) Normalize before compare in test helper (e.g. )\n2) Or enforce LF for golden fixtures via :\n - \n3) Keep output normalization deterministic (trim trailing whitespace/newlines consistently).\n\nIf you want, I can open a follow-up PR/commit with normalization in the test harness.

[alvinreal]

Correction (formatting): Windows failures in PR #88 are likely line-ending normalization issues in golden-file comparisons (CRLF vs LF), not core logic regressions, since Linux/macOS are green and round-trip tests pass. Suggested fix: normalize line endings in test compare helper (replace CRLF with LF before assert), and optionally enforce LF in tests/golden via .gitattributes. I can help with a follow-up patch if needed.