Enable one to progressively build a Kernel with the kernel_builder

[amccaskey]

Setup kernel_builder to start the wrapped FuncOp with a ReturnOp instead of manually adding one at either to_quake or jitCode, and set the insertion point to right before that terminator.

With this change, and smarter caching of the wrapped ExecutionEngine, the kernel_builder can be more flexible and allow progressive building / checking of the internal Quake code.

auto kernel = cudaq::make_kernel();
auto q = kernel.qalloc(2);
kernel.h(q[0]);

// check the kernel as-is
auto state = cudaq::get_state(kernel);
auto counts = cudaq::sample(kernel);

// Continue building the kernel
kernel.x<cudaq::ctrl>(q[0], q[1]);
state = cudaq::get_state(kernel);
counts = cudaq::sample(kernel);

Previously, this would lead to seg-faults or MLIR errors.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.