feat: add Root::parse_bytes for non-UTF8 input

[philiptaron]

Summary

• Add Root::parse_bytes(&[u8]) method that handles non-UTF8 input
• Invalid UTF-8 sequences are replaced with U+FFFD (replacement character)

Background

The C++ Nix parser uses %option 8bit in its flex lexer, which allows it to handle arbitrary byte values without UTF-8 validation. The raw bytes are preserved in string literals.

rnix uses Rowan for its syntax tree, which requires valid UTF-8. This means we cannot preserve arbitrary bytes exactly. Instead, parse_bytes does lossy UTF-8 conversion - invalid sequences become U+FFFD (�).

Behavior comparison:

Input bytes	nix output	rnix parse_bytes output
{ x = "\xff"; }	{ x = "\xff"; } (raw 0xFF preserved)	{ x = "�"; } (U+FFFD replacement)

This is sufficient for most use cases (linting, formatting, analysis) where exact byte preservation isn't required.

Example

// { x = "\xff"; } with raw 0xFF byte (invalid UTF-8)
let bytes: &[u8] = &[0x7b, 0x20, 0x78, 0x20, 0x3d, 0x20, 0x22, 0xff, 0x22, 0x3b, 0x20, 0x7d];
let parse = Root::parse_bytes(bytes);
assert!(parse.errors().is_empty());
// Output text: { x = "�"; }

Test plan

• Added test case non_utf8_can_be_parsed_with_parse_bytes_issue173
• All existing tests pass

Closes #173