Avoid mapping CCDData uncertainty twice

[timbeccue]

Creating a memory-mapped CCDData currently copies its uncertainty array into two temporary files: once in the constructor and again in the uncertainty setter. Only the second mapping is kept as the uncertainty array, so every construction does an unnecessary full-image allocation and copy.

The original CCDData.__init__ in data.py calls _init_array() before assigning the result to self.uncertainty:

self.uncertainty = self._init_array(uncertainty)

That assignment invokes the uncertainty setter, which calls _init_array() again on the already mapped array:

@uncertainty.setter
def uncertainty(self, value: np.array):
    self._validate_array(value)
    self._uncertainty = self._init_array(value)

With memory mapping enabled, each _init_array() call creates a temporary file and copies the array into a new memory map. The constructor therefore creates two copies of the same uncertainty image.

Pass the original array directly to the setter:

- self.uncertainty = self._init_array(uncertainty)
+ self.uncertainty = uncertainty

The setter still validates the shape and creates the memory map, now only once. This removes one temporary file and one full-image copy per CCDData construction while preserving the uncertainty values.