Allure 3 report statistics and UI are incorrect on Windows for parallel runs with retries #925

Describe the bug

Allure 3 can generate an incorrect report on Windows when the input allure-results directory contains parallel/xdist-produced result files that share the same raw uuid(due to reruns) which were run parallely.

In this case, the broken test result exists in allure-results, but the generated Allure 3 report does not show the broken test in the UI, and summary.json/widgets/statistic.json do not include the broken count.

When multiple retry result files have the same uuid, they receive the same internal id and overwrite each other in the store. Since directory iteration order differs between Windows and Linux, the final visible result can become OS-dependent.
The same allure-results can produce the expected report on Linux
but an incorrect report on Windows.

To Reproduce

Python - 3.12.10
Allure - 3.6.2
Node - 24.13.1
WINDOWS 11
Running with pytest-xdist and pytest-rerunfailures

Steps to reproduce the behavior:

1. Clone https://github.com/legion-codes/playwright-pytest-bdd-demo.git
2. uv venv and uv sync to setup
   2.1 playwright install chromium
3. pytest -n 4 --reruns 5 (For my case. Issue visible in lower reruns as well)
4. 1 failure in tests/steps/test_practice_test_automation_table_steps.py::test_filter_beginner_python_courses_with_many_enrollments
5. allure serve reports/allure-results or allure generate reports\allure-results -o reports\allure-report --open to see the report build and wrong count in summary.json

This issue is not present if I use allure2 with JAVA in WINDOWS

Expected behavior
Allure 3 should generate the same report from the same allure-results directory on Windows and Linux.

If a retried test ultimately has a visible latest attempt with status broken, the generated report should show that test as broken in the UI, and the generated statistics should include the broken count, for example:

  "total": 26,
  "passed": 25,
  "retries": 1,
  "broken": 1

Screenshots

Expected Correct Behaviour (Working in LINUX)

Current Behaviour In Windows

Desktop (please complete the following information):

• OS: WINDOWS 11
• Browser Chrome Version 147.0.7727.138
• Version 3.6.2

Additional context
In my reproduction, making the internal result id unique per result file, for example by including context.metadata.originalFileName, makes the broken test appear correctly in the report.

id: md5(${raw.uuid || randomUUID()}:${context.metadata?.originalFileName ?? ""})

context.metadata.originalFileName is already provided by the Allure2 reader here:

await visitor.visitTestResult(dest, { readerId, metadata: { originalFileName } });

I will wait for further Comments and discussion. I can also start by adding a fix for this issue ... if thought process align.