The `tokens` option implicitly enables `parts`, but did not enable
full-pattern scanning. This caused scanning to stop at the first glob
and merge the remaining path segments into the final token.

Include `tokens` in `scanToEnd` and add regression coverage for #62.

Preserve leading and trailing empty segments and return the final
segment for patterns without a path separator.

Track nested and escaped parentheses so that only top-level,
unescaped path separators split the pattern. Keep prefixes out of
the returned parts.

Fixes #58.

fix: scan full pattern when tokens are requested

fix: return complete pattern parts from scan