Parses a TOML v1.0 document into a Document AST.

The parser is a thin layer over Lexer: it consumes the token stream, builds AST nodes, and decodes value text through ValueDecoder. Trivia tokens (Whitespace, Newline, Comment) are not skipped — they are folded into the surrounding node's raw_* fields so that Document#to_toml produces byte-identical output for unmodified documents.

Scope of the current MVP

Implemented:

• Top-level key = value lines
• [a.b.c] standard table headers
• [[a.b]] array-of-tables headers
• Bare keys, quoted keys (basic and literal), dotted keys
• Values: strings (basic, multi-line basic, literal, multi-line literal), integers (decimal, hex, octal, binary, with optional sign and _ digit separators), floats (including inf and nan), booleans
• Standalone comment lines and blank lines

Not yet implemented (planned for the next iteration):

• Datetimes (RFC 3339 offset, local datetime, local date, local time)
• Arrays [1, 2, 3]
• Inline tables { key = "val" }
• Validation that the same key is not redefined and that an already-defined inline table is not extended later