[MLIR] Pass hostShared flag in gpu.alloc op to runtime wrappers (#66401)

This PR is a breakdown of the big PR
https://github.com/llvm/llvm-project/pull/65539 which enables intel gpu
integration. In this PR we pass hostShared flag to runtime wrappers
(required by SyclRuntimeWrappers which will come in subsequent PR) to
indicate if the allocation is done on host shared gpu memory or device
only memory.