[SYCL] Clean up handler.hpp transitive include chain

[koparasy]

Drop dead/unnecessary includes from handler.hpp and its dependency chain to reduce compile-time overhead and improve header hygiene. No ABI or API changes.

Changes:

• Split tls_code_loc_t and code_location into a new sycl/detail/code_location.hpp; common.hpp no longer carries them. Files that relied on the transitive pull now include code_location.hpp directly (accessor.hpp, buffer.hpp, buffer_impl.{hpp,cpp}, common.cpp, usm_impl.hpp, layout_tls_code_loc_t.cpp).
• Drop 14 unused includes from cg_types.hpp; kept only kernel_handler.hpp plus required std headers.
• Drop unused graph.hpp and compile_time_kernel_info.hpp from kernel_launch_helper.hpp.
• Replace device.hpp with class fwd-decl in range_rounding.hpp; only const device& is used.
• Drop unused includes from handler.hpp: cluster_group_prop.hpp, use_root_sync_prop.hpp, detail/ur.hpp, detail/string.hpp, .
• Replace common.hpp with code_location.hpp in handler.hpp.
• Add explicit host_profiling_info.hpp to host_task.hpp (was transitive).

E2E test fixes:

• Tests using parallel_for_work_item relied on cg_types.hpp transitively pulling h_item.hpp. Add explicit <sycl/h_item.hpp> to: HierPar/hier_par_basic.cpp, HierPar/hier_par_indirect.cpp, HierPar/Inputs/hier_par_wgscope_impl.hpp, InorderQueue/in_order_kernels.cpp, Properties/cache_config.cpp, MemorySanitizer/local/parallel_for_work_group.cpp.