feat: add retry mechanism for predict requests

[CarltonHere]

The vllm-server may encounter a small probability 502 error when inferring certain PDFs with a large number of pages, resulting in abnormal interruption of inference. This PR adds a retry mechanism to increase stability during the inference process

Before modification

uv run mineru -p F:\demo.pdf -o .\output\ -b vlm-http-client -u http://***:30000
2025-10-20 15:39:31.089 | INFO     | mineru.backend.vlm.vlm_analyze:get_model:128 - get http-client predictor cost: 0.85s
Two Step Extraction:  75%|████████████████████████████████████████████▏              | 315/421 [06:06<02:03,  1.16s/it]

unhandled exception during asyncio.run() shutdown
task: <Task finished name='Task-4412' coro=<gather_tasks.<locals>.indexed() done, defined at F:\mineru\.venv\Lib\site-packages\mineru_vl_utils\vlm_client\utils.py:96> exception=ReadError('')>
Traceback (most recent call last):
...
  File "F:\mineru\.venv\Lib\site-packages\mineru_vl_utils\vlm_client\http_client.py", line 447, in aio_predict
    response_data = self.get_response_data(response)
                    │    │                 └ <Response [502 Bad Gateway]>
                    │    └ <function HttpVlmClient.get_response_data at 0x000001F8A465C4A0>
                    └ <mineru_vl_utils.vlm_client.http_client.HttpVlmClient object at 0x000001F8A454EF90>
  File "F:\mineru\.venv\Lib\site-packages\mineru_vl_utils\vlm_client\http_client.py", line 218, in get_response_data
    raise ServerError(
          └ <class 'mineru_vl_utils.vlm_client.base_client.ServerError'>
...
2025-10-20 15:45:50.003 | ERROR    | mineru.cli.client:parse_doc:201 - Unexpected status code: [502], response body: ***

After modification

uv run mineru -p F:\demo.pdf -o .\output\ -b vlm-http-client -u http://***:30000
2025-10-20 16:27:15.123 | INFO     | mineru.backend.vlm.vlm_analyze:get_model:128 - get http-client predictor cost: 0.67s
Two Step Extraction:  49%|██████████████████████████████████████████████████████▎                                                        | 206/421 [05:33<00:24,  8.63it/s]

#  If debug is enabled, error messages will be printed
Attempt 1/3 failed: Unexpected status code: [502], response body: ***

Two Step Extraction: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████| 421/421 [07:09<00:00,  1.02s/it]
2025-10-20 16:35:04.875 | INFO     | mineru.cli.common:_process_output:155 - local output dir is .\output\demo\vlm

[Copilot]

Pull Request Overview

This PR adds a retry mechanism to the HTTP VLM client to handle intermittent 502 errors from the vllm-server during PDF inference, improving stability by automatically retrying failed requests up to 3 times before giving up.

• Implements retry logic with configurable max retries (set to 3) for both synchronous and asynchronous predict methods
• Adds error logging for failed attempts when debug mode is enabled
• Prevents inference interruption from temporary server errors

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[myhloli]

Thank you for submitting your code. After carefully evaluating this request, we have implemented the request retry functionality using the httpx-retries library. Therefore, we are closing this PR. We look forward to your future contributions!