Recognize complete BibTeX block starts #583

Problem

The splitter recognizes a block start with @\w* followed only by spaces or tabs before the opening delimiter. This is narrower than the BibTeX data-language structure in two important ways:

• line endings and other ordinary whitespace may separate an entry type from its outer { or ( delimiter; and
• entry-type identifiers may contain valid non-word punctuation such as -, :, or /.

Candidates outside the regular expression are currently retained as implicit comments rather than becoming parsed blocks or explicit failures. That makes a structural input disappear from the bibliography model without any parse failure.

Proposed behavior

Recognize the complete block-type token up to whitespace or a reserved BibTeX syntax character, locate the following standard outer delimiter across whitespace, and use the delimiter position rather than assuming it immediately follows the regex match.

Add a matrix requiring every recognized candidate to become either:

• a parsed block for the currently supported curly-brace form; or
• an explicit failed block for the currently unsupported parenthesis form.

This work builds on the explicit parenthesis-failure handling in #533 / PR #571 and exact special-command matching in #568 / PR #574. The implementation PR should retain those as separate prerequisite commits so the broader parser change remains reviewable.