Honor sort_keys for parsed TOML documents

[ShipItAndPray]

Summary

• honor sort_keys=True when dumping parsed TOMLDocument / container values
• keep the existing fast path for unsorted container dumps
• add regression coverage for loads() + dumps(sort_keys=True)

Closes #466.

Validation

Autoresearch-style binary gates:

• E1 baseline repro exists: tomlkit.dumps(tomlkit.loads('zzz = 1\naaa = "foo"'), sort_keys=True) preserved the original order on master
• E2 regression expectation fails on baseline: the new test expects sorted output for a parsed document
• E3 regression passes after fix: parsed documents now dump as 'aaa = "foo"\nzzz = 1\n'
• E4 targeted tests pass:
  • python3 -m pytest tests/test_api.py -k 'sorted_keys or raw_dict_can_be_dumped or mapping_types_can_be_dumped'
• E5 broader nearby suite passes:
  • python3 -m pytest tests/test_api.py

Notes

The fix only changes the sort_keys=True container path. Unsorted container dumps still use the existing direct as_string() fast path.