fix: ignore natural-language percentages as placeholders

[vitormattos]

Summary

Fixes false-positive printf placeholder detection in natural-language percentages.

The placeholder regex currently accepts a literal space as a printf flag. That makes ordinary prose such as:

• 22% growth look like a %g placeholder;
• 22% e deve... look like a %e placeholder.

This causes valid translations to fail placeholder validation.

Changes

• stop treating a plain space after % as a printf flag;
• add regression coverage for percentages followed by words;
• preserve existing support for actual printf placeholders, widths, precision, positional arguments and escaped percent signs.

This was exposed by the pt-BR migration in LibreSign/site#506.