…usage

…ected arguments matches

… be modified

…nated string

…riables

…tring

…presentation

…ython objects

…h stricter tolerances

…cks have passed

…ccepted

…turn arguments

…to keep reference

Somehow, this small refactoring took too much time, and I still do not quite understand why.
For some reason, the reference to the sysconfig module must be preserved.

…r backward compatibility

… strides)

…ruct

…ting it each time

…nagement

…annot be mutable

…object

For some reason, this problem has never manifested itself before,
although it is clear from the code that I have the same fix
for a Python OIF_CALLBACK object.
When I have started to work on the `optim` interface, this reference deficit
started to happen quite often, hence the fix with `Py_INCREF`.
Reference deficit is when a `PyObject *` does not have its refcount
incremented enough times.

…ointers

If the user passes to `oif_create_array_f64` for the `dimensions` parameter
a stack-allocated array, then it can easily happen that in the created array
the field `dimensions` will be a dangling pointer at some point.
To prevent this, we copy the passed array so that it stays with the array
during its lifetime.

…ensions` field

[misc] Debug CI

…imensions

I really do not know how this thing were working before:
Python was using 32-bit integers while C and Julia 64-bit integers to represent
dimensions of an array.
How they were able to convert them, remains unknown.