1.7.0

Changed

• Confidence assessment uses the parser's configured honorific connectors, so unattributed German or custom connectors no longer hide title/name ambiguity. Parsed name fields are unchanged.
• Clarify whitespace normalization and the retained interior-tab exception for an empty whitespace set in the README. Keep the tab exception in one place without repeating the older caveat.
• Clarify in the README that invalid UTF-8 is scrubbed with the process-wide mb_substitute_character() setting, which the parser does not change, and that a custom setMappers() list still applies to unsplit no-comma input under setSurnameFirst(true) (Kim, Dr. Kim). Parsing behavior is unchanged.

Fixed

• Explicit dot-separated initials no longer become surname particles in mixed-case names: John D.E. Smith keeps initials D E and surname Smith. Undotted particles, uniform-uppercase ambiguity handling, and configured combined-initial limits are unchanged.
• Nickname-delimiter sanitization no longer copies every valid configured pair before keeping the first 32. Large configurations use bounded temporary storage while retaining validation of all entries.
• Surname particles remain attached after given-name initials: J. van Gogh now keeps first name J. and surname van Gogh, rather than promoting van to the first name. This also applies with a salutation, nickname, or trailing credential. See the parser regression cases.
• Numeric symmetric nickname delimiters now obey token boundaries when splitting names and shield nickname commas during standalone confidence assessment.
• The input byte limit is checked before UTF-8 scrubbing as well as after it, so malformed input cannot bypass the limit when replacement removes or shortens invalid sequences.