Because the name of the kernel is calculated in the constructor,
any child class that had overriden the kernel name, will be
ignored.
By setting the kernel name in the child class, we can avoid having
to execute the regex to calculate the name upon every construction
of a Kernel.
A test (and a kernel fixture) is added to prove that the override
works correctly.
Note: the Kernel API has not been touched, so there should be no
issues with BC.