Add gap and unicode parse fixtures

[primeBeri]

PR summary — Add gap and unicode parse fixtures

Summary

Adds 30 new parse assertions and one document pair to the shared fixture suite,
derived from the Huml.Net (.NET reference implementation) test suite. All cases
are implementation-agnostic: they only assert whether a given input parses
successfully or throws a parse error, with no dependency on token types, AST
structure, error messages, or runtime behaviour.

Changes

• assertions/mixed.json — 30 new entries (12 gaps + 18 unicode)
• documents/unicode.huml — new document fixture
• documents/unicode.json — expected parse output for unicode.huml

Categories

Gaps (v0.2)

Parse behaviours that were exercised by Huml.Net's test suite but absent from
the shared suite. Covers:

• Case-sensitivity of keywords: TRUE, FALSE, NULL are accepted (same as
  lowercase)
• Tab character at line start is an error
• Quoted key with a non-string value ("my-key": 42)
• Root-level scalar types not previously covered: float (3.14), nan, inf,
  hex (0xFF)
• Multiline list with integer items
• Bare key:: with no following content is an error
• Quoted key containing a colon ("a:b")

Unicode (v0.2)

Covers the boundary between bare keys (ASCII-only identifiers) and quoted keys,
and validates that non-ASCII content is accepted inside quoted strings and
quoted keys:

• Bare non-ASCII keys (Arabic, Chinese, Cyrillic, Devanagari, emoji) are errors
• Quoted non-ASCII keys (same scripts) are accepted
• Non-ASCII string values (Arabic, Hebrew, Chinese, Korean, emoji) are accepted
• Mixed LTR/RTL strings are accepted
• RTL mark (U+200F) and LTR mark (U+200E) inside strings are accepted

Notes

• No name collisions with existing entries in mixed.json
• The closest existing entry is ambiguous_empty_vector_space (key:: # comment);
  the new ambiguous_empty_vector_bare (key::) tests a distinct code path
• All non-ASCII and invisible codepoints in the JSON source use \uXXXX escapes
  to keep the files ASCII-safe