[Numpy] Advanced Indexing Bug

[sxjscience]

@AntiZpvoh Would you still have time to take a look?

Reproducible example:

import mxnet as mx
import numpy as np

mx.npx.set_np()
batch_size = 4
seq_length = 16
num_sel_positions = 5
feature_shape = (16, 32)
data = mx.np.random.uniform(-1, 1, (batch_size, seq_length) + feature_shape, dtype=np.float32)
positions = mx.np.random.randint(0, seq_length, (batch_size, num_sel_positions), dtype=np.int32)
out_np = data.asnumpy()[np.expand_dims(np.arange(data.shape[0]).astype(np.int32),
                                       axis=1),
                        positions.asnumpy()]
out_mx = data[mx.np.expand_dims(mx.npx.arange_like(data, axis=0).astype(np.int32), axis=1), positions]
out_mx.asnumpy()

Error:

MXNetError: Traceback (most recent call last):
  File "../include/mxnet/./tensor_blob.h", line 311
TBlob.get_with_shape: Check failed: this->shape_.Size() == static_cast<size_t>(shape.Size()) (20 vs. 4) : new and old shape do not match total elements