Support more of abstract array interface

Currently `similar` will return an array of a different type:

```python
from juliacall import Main as jl
import numpy as np

x = np.random.randn(5)
y = jl.similar(x)

print(jl.typeof(x))
# PyArray{Float64, 1, true, true, Float64}
print(jl.typeof(y))
# Vector{Float64}
```

This can introduce some type instabilities in libraries due to the assumption that container type is preserved by `similar`.

I guess we just need the "optional methods" from here: https://docs.julialang.org/en/v1/manual/interfaces/#man-interface-array

No implementation file or test is named. Reproduce the Python/Julia `similar` example from the issue, then read the linked Julia array-interface documentation and trace the existing `similar` handling in PythonCall.jl. Done means the relevant optional array-interface methods are supported and the example preserves the expected container type.

索引モデルが issue の本文から書いたものです。