StringGetOwnProperty does not produce the same strings that the lexer produces

[raskad]

Describe the bug
The StringGetOwnProperty operation used when indexing a string ("abc"[1]) does not produce the same string that the lexer produces. This matters for UTF16 surrogates.

Currently the lexer uses String::from_utf16_lossy (see UTF16CodeUnitsBuffer::to_string_lossy) and StringGetOwnProperty uses char::from_u32(u32::from(utf16_val)).map_or_else(|| Value::from(format!("\\u{:x}", utf16_val)), Value::from) (ignore the conversion to Value).

To Reproduce
This JavaScript code reproduces the issue:

if ('𝌆'[1] == '\udf06') {
    console.log(true)
} else {
    console.log(false)
}

Expected behavior
Both should produce the same strings, because the spec expects them to be equal.

Build environment:

• OS: Linux
• Version: fff3974
• Target triple: x86_64-unknown-linux-gnu
• Rustc version: rustc 1.53.0 (53cb7b09b 2021-06-17)