analyzer: add pluggable tokenizer and basic CJK (code-point) support

[amirHdev]

Adds a minimal, backwards-compatible path for CJK search support by making tokenization configurable and providing a built-in Unicode code-point tokenizer.

What changed

• Added TokenizerFunc to AnalyzerConfig for custom tokenization.
• Added UnicodeCodePointTokenizer for rune/code-point tokenization.
• Added DefaultCJKConfig():
  • code-point tokenizer
  • MinTokenLength = 1
  • disables English stopword filtering and stemming
• Added NewInvertedIndexWithAnalyzer(config) to construct an index with custom analyzer behavior.
• Routed indexing and query-time analysis through the index analyzer config (indexing, BM25/proximity ranking, query builder term/phrase).
• Added tests covering:
  • tokenizer output for Unicode/CJK text
  • BM25 match for "你好，世界" queried with "好"
  • QueryBuilder term query honoring the index analyzer config

Why

Current behavior handles Unicode characters but tokenizes CJK text as whole words (e.g. "你好"), so single-character queries like "好" do not match. This PR adds a simple tokenizer strategy that enables basic CJK matching without changing default English behavior.

Notes / limitations

• This is a basic code-point tokenizer, not a full Chinese/Japanese segmentation solution.
• It improves correctness for single-character matching and provides an extension point for future tokenizers (e.g. bigram or dictionary-based segmentation).

Example

idx := blaze.NewInvertedIndexWithAnalyzer(blaze.DefaultCJKConfig())
idx.Index(1, "你好，世界")
matches := idx.RankBM25("好", 10)