Cannot resize or concat NDArray created with shape=0

[Jacc4224]

Describe the bug
When creating a blosc2.NDArray with shape=0, subsequent resize() and concat() operations fail or behave unexpectedly.

To Reproduce
Steps to reproduce the behavior:

import blosc2
import numpy as np

# Define dtype
dt = np.int64

# Create empty array (shape=0)
my_array = blosc2.zeros(shape=0, dtype=dt)

# Resize causes crash
my_array.resize((1,))

Output error:
Process finished with exit code 136 (interrupted by signal 8:SIGFPE)

Expected behavior
Should be able to resize array from shape=0 to any size, and concatenate empty arrays with non-empty arrays.

Desktop (please complete the following information):

• Python 3.12.4
• Blosc2 version: 4.0.0-b2.dev0
• OS: Ubuntu 24.04.3 LTS

Additional context
This issue affects the implementation of dynamically growing columnar tables where columns need to start empty and grow as data is appended. The crash (SIGFPE) suggests a division by zero or arithmetic exception at the C level when handling zero-sized arrays.

[lshaw8317]

I accept this with a slight modification - my_array = blosc2.zeros(shape=0, dtype=dt, chunks=(100,)) should allow resizing, but my_array = blosc2.zeros(shape=0, dtype=dt) should set chunks = (0,) and then fail on resizing. One could possibly add an exception for the 1D case, since then one may unambiguously calculate an optimal chunkshape, but I am not sure.

[FrancescAlted]

Ok. My proposal is that, when shape=(0,) (or maybe even (1,)), to enable specific paths in compute_chunks_blocks() to provide decent values for chunks and blocks. If this still creates issues with (0,), then raise an explicit error telling the user that in that case, chunks param is explicitly required.