[Question] af.lapack.qr request 7TB of RAM

Hello!
I was trying to compute a QR decomposition on the GPU using OpenCL for a matrix like this:
```
shape: (1000000, 40)
dtype: float64
bytes: 320 MB
```
But the function `af.lapack.qr` returns an `out_of_memory ` error:
```
RuntimeError: In function nativeAlloc
In file src/backend/opencl/memory.cpp:197
Failed to allocate device memory of size 7.64 TB
```
It is normal to allocate such a ridiculous amount of memory?
CUPY can solve this QR with reasonable amounts of RAM

Machine info:
```
CPU: 2 x Intel Xeon Platinum 8160
RAM. 512 GB
GPU: NVIDIA V100
```

Start with the af.lapack.qr call and the allocation path reported at src/backend/opencl/memory.cpp:197. Reproduce the 1,000,000 by 40 float64 QR case on the stated GPU, then trace why nativeAlloc requests 7.64 TB and compare the allocation behavior with the expected QR workspace; done means the excessive allocation is explained and a focused fix or confirmed limitation is documented.

索引モデルが issue の本文から書いたものです。