Reject keys that are invalid for an explicit bare or literal key type

[choudhryfrompak]

Summary

SingleKey only escapes basic keys. If you pass t=KeyType.Bare or t=KeyType.Literal yourself, the key text goes between the delimiters as is. A newline, =, a quote or a dot in the key then changes the structure of the dumped document:

import tomlkit
from tomlkit.items import KeyType, SingleKey

doc = tomlkit.document()
doc.add(SingleKey('user = "x"\nadmin', KeyType.Bare), True)
doc.add(SingleKey("x' = 1\n[evil]\n'y", KeyType.Literal), 2)
print(tomlkit.dumps(doc))
# user = "x"
# admin = true
# 'x' = 1
# [evil]
# 'y' = 2

That parses back as {'user': 'x', 'admin': True, 'x': 1, 'evil': {'y': 2}}, not as two keys. Same on 0.15.1 from PyPI.

With this change SingleKey raises ValueError when an explicit type can't represent the key. A bare key has to be non-empty and use only A-Za-z0-9_-. A literal key can't contain ' or a control character other than tab, which is the rule StringType.SLL.invalid_sequences already has for literal strings. Basic keys were already escaped and haven't changed. Keys from the parser pass original and skip the check, and so do keys whose type is picked automatically (tomlkit.key(), doc["..."], table.add("...")). I moved the bare character check into a small _is_bare_key() helper so the automatic choice and the new check use the same rule.

This can break code that used KeyType.Bare to write a dotted header on purpose, e.g. SingleKey("tool.poetry", t=KeyType.Bare). I found that in one small project on GitHub. It already gave an inconsistent document (doc.unwrap() has a "tool.poetry" key while the output parses as nested tables), and table(is_super_table=True) gives the same [tool.poetry] header properly. If you'd rather keep dots allowed for bare keys, I can relax it.

Tests are in tests/test_items.py: invalid bare and literal keys raise, and valid keys of each explicit type round-trip through parse. The 8 rejection cases fail on master. The full suite passes with the toml-test submodule checked out (1080 passed, 1068 before), and ruff and ruff-format are clean on the changed files.