Fix regression in String.from_raw and improve documentation - #658

AI-assisted validation at 386786c7a48c on Python 3.12.14: I checked basic/literal and single/multiline strings with backslashes, a tab, Unicode text, and ordinary text. With escaping enabled, all 16 combinations round-tripped through tomllib with their input value intact. For multiline strings beginning with one or two LFs, I separately verified that escape=False emits the supplied body verbatim, while the default mode adds the trim-protection newline. The 79 string-focused API/item tests passed.

That separation is the behavior I would preserve in regression coverage: verbatim mode may intentionally give a parsed value different from the raw body because TOML trims an opening newline. It should not silently gain the extra newline needed by the default value-preserving mode.