[FEATURE] add Lxmert model

[zhtmike]

What does this PR do?

Fixes # (issue)

Adds # (feature)
Add Lxmert model

>>> from transformers import AutoTokenizer
>>> from mindone.transformers import LxmertForQuestionAnswering
>>> import mindspore as ms

>>> tokenizer = AutoTokenizer.from_pretrained("unc-nlp/lxmert-base-uncased", revision="refs/pr/3")
>>> model = LxmertForQuestionAnswering.from_pretrained("unc-nlp/lxmert-base-uncased", revision="refs/pr/3")

>>> question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet"

>>> inputs = tokenizer(question, text, return_tensors="np")
>>> for k, v in inputs.items():
...     inputs[k] = ms.tensor(v)

>>> outputs = model(**inputs)

>>> answer_start_index = outputs.start_logits.argmax()
>>> answer_end_index = outputs.end_logits.argmax()

>>> predict_answer_tokens = inputs.input_ids[0, answer_start_index : answer_end_index + 1]
>>> tokenizer.decode(predict_answer_tokens, skip_special_tokens=True)

>>> # target is "nice puppet"
>>> target_start_index = ms.tensor([14])
>>> target_end_index = ms.tensor([15])

>>> outputs = model(**inputs, start_positions=target_start_index, end_positions=target_end_index)
>>> loss = outputs.loss
>>> round(loss.item(), 2)

will encounter same error as huggingface/transformers#7266 (comment)

need to migrate demo to validate end-to-end accuracy: https://github.com/huggingface/transformers-research-projects/tree/main/lxmert

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
  documentation guidelines
• Did you build and run the code without any errors?
• Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@xxx

[zhtmike]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces the Lxmert model, a multi-modal transformer for language and vision, ported from Hugging Face's PyTorch implementation to MindSpore. The core model logic and auto-model registration are included. However, the review identified several critical issues stemming from the direct porting of PyTorch-specific code, which will lead to runtime errors. These include incorrect weight initialization and tensor manipulation syntax. Additionally, the accompanying tests are incomplete, covering only the base model and not the heads for pre-training or question answering. I have provided detailed comments and suggestions to address these issues, fix the bugs, and enhance the overall code quality and test coverage.