[core][ndarray] Fix shape reconstruction logic in NDArray.__getitem__()

[josiahls]

Caused by:

ValueError: Number of dimensions must be positive, got 0. Provide at least one shape dimension. [at NDArrayShape.__init__(shape: List[Int])]

on

print(array[0, 0])

Additionally added a TODO since the error message was otherwise extremely vague.

---

Updates during review by @forfudan:

index_type_list.append(IndexTypeInfo(is_integer=True)), without which the mixture of slices and integers would cause integers to be missed in the record.

However, this PR is not sufficient to address the problem. There are some other bugs in the code related to shape reconstruction logic, which will be even exposued after adding the index_type_list.

Here are some further fixes:

1. Fixed is_integer branch: The old code new_shape.append(narr.shape[i]) was wrong in two ways — it used i (position in index_type_list) instead of a proper dimension counter, and it added the dimension to the output shape when integer indexing should remove it (dimension reduction). Changed to just advance narr_dim without appending (skip it).

2. Fixed missing trailing dimensions: When index_type_list has fewer entries than narr's dimensions (e.g., b[Slice(0,2)] on a 2D array), remaining dimensions were silently dropped. Added a while narr_dim < ndims loop to preserve them.

3. Fixed ellipsis dimension counting: Changed from self.ndim - len(index_type_list) + 1 (which incorrectly assumes every non-ellipsis entry consumes a dimension — NewAxis doesn't) to counting only is_slice and is_integer entries.

4. Removed stale TODO from the original PR: # TODO: Shouldn't we assert len(shape_size) == len(slice_list)? referenced a nonexistent variable shape_size.

[josiahls]

Note it looks like a lot of unit tests in NuMojo/tests/core/test_array_indexing_and_slicing.mojo are commented out, which might explain why this got through.

[forfudan]

@josiahls, thank you for the PR.

Your PR catched a missing index_type_list.append(IndexTypeInfo(is_integer=True)), without which the mixture of slices and integers would cause integers to be missed in the record.

However, this PR is not sufficient to address the problem. There are some other bugs in the code related to shape reconstruction logic, which will be even exposued after adding the index_type_list.

So I made some other fixes on top of your PR, changed upon yours, and update the description.

---

@shivasankarka, I am adding tests for this case, please let me handle the approval for this PR. Thanks :D

[forfudan]

Approved with some further fixes, need to also add some tests in another PR.

[shivasankarka]

@josiahls @forfudan Thanks for the fix :)