fix: skip a UTF-8 BOM in the fast parser

The character scanner used by { fast: true } skipped only spaces, tabs and
newlines between entries, so a leading BOM was treated as the start of a key.
It matches no key character, the line was discarded as invalid, and the first
variable in the file silently went missing.

The classic regex parser consumes the BOM as part of its leading \s*, so the
two parsers disagreed on any .env saved with a UTF-8 signature — the default
encoding for PowerShell redirects and Windows Notepad.