GLQL query returns incorrect results when using Elasticsearch vs database

Summary

GLQL queries with label filters return different (incorrect) results when using Elasticsearch compared to direct database queries. A specific query for Plan Stage Roadmap epics returns only 1 result via ES but 10 results when bypassing ES with ?useES=false.

Note

To reproduce it on .com, you need to enable the glql_es_integration FF for your user via chatops

Steps to reproduce

1. Create a GLQL query with label filters:

   ```glql
   display: table
   title: Plan Stage Roadmap
   fields: labels("group::*") as "Group", due, title, state, milestone, assignees
   query: group = "gitlab-org" AND type = Epic AND label = ~"devops::plan" and label = ~"Plan Roadmap" 
   ```

2. Embed this query in a wiki page or issue note
3. View the results (should show only 2 epics)
4. Add ?useES=false parameter to the page URL
5. Compare results (should now show 11 epics)

What is the current bug behavior?

• GLQL query returns only 1 epics when using Elasticsearch
• The same query returns 10 epics when bypassing Elasticsearch (?useES=false)
• Regular issue listing pages show the correct 10 results

What is the expected correct behavior?

• GLQL query should return the same results whether using Elasticsearch or database queries
• Both should return 10 epics matching the specified criteria
• Results should be consistent with regular GitLab issue/epic listing functionality

Relevant logs and/or screenshots

Test URLs demonstrating the issue:

• With ES (shows 1 results): https://gitlab.com/gitlab-org/plan-stage/test-project/-/issues/265
• Without ES (shows 10 results): https://gitlab.com/gitlab-org/plan-stage/test-project/-/issues/265?useES=false

Possible fixes

The issue likely lies in one of these areas:

• GLQL to Elasticsearch query translation, particularly for label matching syntax (~"label")
• Label field mapping or analyzer configuration in Elasticsearch
• Boolean logic interpretation differences between ES and database queries
• Case sensitivity or special character handling in label searches

Investigation should focus on comparing the actual ES query generated vs the database query for the same GLQL input.