Add Image.array setter (#1272)

[rmjarvis]

@sidneymau ran into a weird edge case in a Jupyter notebook when doing im.array /= 10 raised an AttributeError (that array doesn't have a setter), but performed the operation nonetheless.

This seems to be a bug in either Python or numpy. Numpy overloads the inplace arithmetic operations so ar /= 10 is allowed and works in place. But Python seems to think that im.array /= 10 would necessarily require a setter, so it raises an AttributeError. This so far is fine, but then Python calls the numpy __idiv__ operation anyway, so the operation succeeds along with an AttributeError being raised.

It was at least confusing for Sid, since a Jupyter session doesn't die when that happens, so the state of the array seemed inconsistent with the error messaging.

Anyway, this PR adds a setter, which lets im.array = blah work the same way as im.array[:] = blah, with the slight exception that we are careful about rounding things to the nearest integer if we are assigning to integer arrays rather than truncating as numpy would want to do. (This is consistent with how GalSim does other arithmetic on integer Images.)

Note, this replaces Sid's PR #1276, which I wasn't happy with. This implementation never changes the underlying array object. Just lets things be assigned to it as appropriate.

[sidneymau]

Just clarifying that this behavior was not specific to jupyter notebooks but also occurred in any interactive (including pdb/breakpoint) session from what I recall

[sidneymau]

Everything looks ok to me and tests pass.