Update sharded addrgen and CCL ops to work with custom core list

[SeanNijjar]

A recent feature was added to enable L1 sharded tensors to be sharded across custom core lists/ranges. This issue tracks the enablement of that feature in sharded addrgen, which is used by CCL operations.

sharded addrgens can be found here:
ttnn/cpp/ttnn/operations/ccl/shared_with_host/sharded_tensor_addr_gen.hpp

To support this feature mainly three changes will be needed:

1. A new WorkerToNocCoordLookup will need to be implemented (e.g. ArbitraryShardCoreListCoordLookup)

• Please add corresponding tests to tests/ttnn/unit_tests/gtests/ccl/test_ccl_tensor_slicers.cpp - should also simplify initial bringup and testing

2. kernel side: the kernel will need to have a different codepath for instantiating this new type of coord lookup - based off of a new CT arg we pass in

• Runtime args will accordingly need to be interpreted in a different way
  • Luckily we are already consuming RT args for coord lookup today for harvesting, so the general type of codepath already exists

3. Update RT/CT arg generators: emit_address_generator_runtime_args for this mode (to be detected from the mem config)

• For now I think we should leave CT args alone since we will otherwise need to update many other kernels. This could be done but I think a more comprehensive update of CT args makes sense to wait for this feature to be completed: Improve Compile Time Args to Make Them Programmatically Accessible (TMP) #16040

[SeanNijjar]

tracking moved to #16608