fix: add missing signature

[henryiii]

All the other array constructors have an optional handle, but this one was missing.

[YannickJadoul]

I was about to say "Isn't that because it's a constructor from buffer_info?", but apparently buffer_info doesn't contain a base either? I'm not a buffer-protocol expert, though, so I don't know if it should be in there. If not, this seems fine.

Maybe do the same for the array_t constructor, though?

[henryiii]

FYI, here's where I ran into this:

https://github.com/scikit-hep/boost-histogram/blob/026e5e5c3ec709083fadc3ead61408d030d03d64/include/bh_python/register_histogram.hpp#L125-L132

[henryiii]

Someone told me when I found this a while back it was probably just an oversight, but I don't know where that conversation was.

[henryiii]

By the way, buffer info is information about the data, but it does not contain ownership information. The handle contains ownership information, so you can be sure the owning object is not destroyed while the array is still present. The other constructors support this, it's just missing from the direct buffer info constructor.

[YannickJadoul]

Alright; Py_buffer does contain a object field, but py::buffer_info doesn't. I don't know enough about the buffer protocol to assess the details of what should do what, but adding these optional base arguments to the signature seems fine to me!

[rwgk]

Looks good to me!
I spent a few minutes looking in the nump.h sources, this change seems super safe to me.