The `filter_by` is not working as expected for hybrid search

Bug Description

I expect the query to have 0 hit, nothing matched, but Typesense gives me result.

Reproduction Steps

#!/bin/bash

# sematic search with auto embedding
# filter by field

QUERY="nonexistent"
CHUNK=300

curl -s -X POST "http://172.16.8.107:8108/multi_search" \
  -H "Content-Type: application/json" \
  -H "X-Typesense-Api-Key: typesense" \
  -d '{
    "searches": [
      {
        "collection": "stock_reports_cn",
        "q": "'${QUERY}'",
        "query_by": "title,content,name,code,category,institution,embedding",
        "per_page": 2,
        "filter_by": "name:='宁德时代'",
        "highlight_fields": "title,content,institution",
        "highlight_affix_num_tokens": '${CHUNK}',
        "sort_by": "_text_match:desc, date:desc",
        "exclude_fields": "path,content,embedding"
      },
      {
        "collection": "stock_notices_cn",
        "q": "'${QUERY}'",
        "query_by": "title,content,name,code,category,embedding",
        "vector_query": "embedding:([], k:100, distance_threshold:0.2, alpha: 0.7)",
        "per_page": 2,
        "filter_by": "name:='宁德时代'",
        "highlight_fields": "title,content",
        "highlight_affix_num_tokens": '${CHUNK}',
        "sort_by": "_text_match:desc, date:desc",
        "exclude_fields": "path,content,embedding"
      }
    ]
  }'

Expected vs Actual

Expected behavior
nothing matched.

Actual behavior
Typesense gives me result.

{
  "results": [
    {
      "facet_counts": [],
      "found": 100,
      "hits": [
        {
          "document": {
            "category": "stock",
            "code": "300750",
            "created": 1773733034,
            "date": 1742256000,
            "id": "AP202503181644493316",
            "institution": "交银国际证券有限公司",
            "name": "宁德时代",
            "title": "全球扩产提速，盈利能力保持韧性；维持买入"
          },
          "highlight": {},
          "highlights": [],
          "hybrid_search_info": {
            "rank_fusion_score": 0.30000001192092896
          },
          "text_match": 0,
          "text_match_info": {
            "best_field_score": "0",
            "best_field_weight": 0,
            "fields_matched": 0,
            "num_tokens_dropped": 1,
            "score": "0",
            "tokens_matched": 0,
            "typo_prefix_score": 255
          },
          "vector_distance": 0.2201164960861206
        },
        {
          "document": {
            "category": "stock",
            "code": "300750",
            "created": 1773739651,
            "date": 1729555200,
            "id": "AP202410221640410402",
            "institution": "中银国际证券股份有限公司",
            "name": "宁德时代",
            "title": "盈利能力提升，海外进展顺利"
          },
          "highlight": {},
          "highlights": [],
          "hybrid_search_info": {
            "rank_fusion_score": 0.15000000596046448
          },
          "text_match": 0,
          "text_match_info": {
            "best_field_score": "0",
            "best_field_weight": 0,
            "fields_matched": 0,
            "num_tokens_dropped": 1,
            "score": "0",
            "tokens_matched": 0,
            "typo_prefix_score": 255
          },
          "vector_distance": 0.2214726209640503
        }
      ],
      "out_of": 119314,
      "page": 1,
      "request_params": {
        "collection_name": "stock_reports_cn",
        "first_q": "nonexistent",
        "per_page": 2,
        "q": "nonexistent"
      },
      "search_cutoff": false,
      "search_time_ms": 164
    },
    {
      "facet_counts": [],
      "found": 0,
      "hits": [],
      "out_of": 55489,
      "page": 1,
      "request_params": {
        "collection_name": "stock_notices_cn",
        "first_q": "nonexistent",
        "per_page": 2,
        "q": "nonexistent"
      },
      "search_cutoff": false,
      "search_time_ms": 78
    }
  ]
}

Environment

• Typesense version: 30.1
• Operating system: Ubuntu 22.04.4 LTS
• Client library & version: rest api

Schema / Configuration

# create research collection with auto-embedding fields
curl -X POST http://172.16.8.107:8108/collections \
  -H "X-TYPESENSE-API-KEY: typesense" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "stock_reports_cn",
    "fields": [
      {"name": "id", "type": "string"},
      {"name": "date", "type": "int64", "sort": true},
      {"name": "code", "type": "string"},
      {"name": "name", "type": "string", "locale": "zh"},
      {"name": "category", "type": "string", "locale": "zh"},
      {"name": "institution", "type": "string", "locale": "zh"},
      {"name": "title", "type": "string", "locale": "zh"},
      {"name": "content", "type": "string", "locale": "zh"},
      {"name": "created", "type": "int64", "sort": true},
      {"name": "embedding", "type": "float[]", "embed": {
            "from": ["title", "content"],
            "model_config": {
                "model_name": "ts/gte-large"
            }
         }
      }
    ],
    "default_sorting_field": "date"
  }'

# create notice collection with auto-embedding fields
curl -X POST http://172.16.8.107:8108/collections \
  -H "X-TYPESENSE-API-KEY: typesense" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "stock_notice_cn",
    "fields": [
      {"name": "id", "type": "string"},
      {"name": "date", "type": "int64", "sort": true},
      {"name": "code", "type": "string"},
      {"name": "name", "type": "string", "locale": "zh"},
      {"name": "category", "type": "string", "locale": "zh"},
      {"name": "title", "type": "string", "locale": "zh"},
      {"name": "content", "type": "string", "locale": "zh"},
      {"name": "created", "type": "int64", "sort": true},
      {"name": "embedding", "type": "float[]", "embed": {
            "from": ["title", "content"],
            "model_config": {
                "model_name": "ts/gte-large"
            }
         }
      }
    ],
    "default_sorting_field": "date"
  }'

Additional Context

The highlights worked as expected, nonexistent keyword should match nothing, but the filter dose not work as expected.

Start by reproducing the reported curl request against the REST multi_search endpoint, comparing the filter_by behavior in the stock_reports_cn hybrid search with the stock_notices_cn vector search. Trace how filter_by is applied alongside vector_query and hybrid ranking; done means the name filter excludes all non-matching results and a regression test covers this case.