Fix: RandomSamplingNegPos forget to remove gt_ignore_flags

[jiangtann]

在https://github.com/open-mmlab/mmdetection/blob/dev-3.x/mmdet/datasets/transforms/formatting.py#L109 中，会根据valid_idx = np.where(results['gt_ignore_flags'] == 0)[0]来获得valid_idx，并通过valid_idx从gt_bboxes中取得有效的bboxes。

在https://github.com/open-mmlab/mmdetection/blob/dev-3.x/mmdet/datasets/transforms/text_transformers.py#L62 中，如果positive labels的token数之和超过了256，会随机删掉一部分gt_bboxes和gt_labels，但是没有对gt_ignore_flags进行相应的处理，导致在PackDetInputs中，通过valid_idx从gt_bboxes中取得有效的bboxes会出现index越界error。

参考RandomCrop中处理gt_ignore_flags的方式对RandomSamplingNegPos进行相应的修改

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}