feat: add vector function support

[Mini256]

Inspired by the pgvector-node package: https://github.com/pgvector/pgvector-node/blob/master/src/kysely/index.js

Simplifies the use of TiDB Serverless Vector Function for kysely developers while providing better code hints.

For example:

db.selectFrom('document_index_chunk')
      .innerJoin('document', 'document_id', 'document.id')
      .select([
        'document_index_chunk.id as document_index_chunk_id',
        'document_id',
        'document_index_chunk.text_content',
        'document_index_chunk.metadata',
        'source_uri',
        'document.name as source_name',
-        eb => eb(eb => eb.lit<number>(1), '-', eb.fn<number>('vec_cosine_distance', [
-          'embedding',
-          eb => eb.val(vectorToVal(vector))],
-        )).as('score'),
+        (eb) => cosineSimilarity(eb, 'embedding', vector).as('score')
      ])
      .where('staled', '=', 0)
      .where('index_name', '=', eb => eb.val(index))
      .orderBy('score desc')
      .limit(top_k);

[634750802]

Please consider typing safe for column names (not just using string type). It's the major advantage of kysely.