[NFCI][SYCL] More use of CompileTimeKernelInfoTy/DeviceKernelInfo

[aelovikov-intel]

Additional cleanups enabled by #19117.

[adamfidel]

Changes for SYCL Graph LGTM

[aelovikov-intel]

I've forced pushed after rebasing on top of several smaller changes factored out from here. Doing the same via merges would have been much harder.

SYCL Graphs changes in this PR were not affected by the rebase.

[aelovikov-intel]

getKernelName() is better be left to a subsequent PR, IMO. We have

llvm/sycl/include/sycl/detail/kernel_name_str_t.hpp

Lines 17 to 25 in ed3767c

	#ifdef __INTEL_PREVIEW_BREAKING_CHANGES
	using KernelNameStrT = std::string_view;
	using KernelNameStrRefT = std::string_view;
	using ABINeutralKernelNameStrT = detail::string_view;
	#else
	using KernelNameStrT = std::string;
	using KernelNameStrRefT = const std::string &;
	using ABINeutralKernelNameStrT = detail::string;
	#endif

and some of our internal interfaces do expect const std::string & arguments. Creating those from std::string_view would result in otherwise unnecessary memory allocations.

On the other hand, I'd really prefer we do NOT introduce more ugly instances of KernelNameStrRefT so I'd rather keep getKernelName() returning std::string_view and use that when it's fine but leave explicit accesses to DeviceKernelInfo::Name when const std::string & might be needed.

[aelovikov-intel]

Jenkins failure (old RHEL/gcc) seems to have been caused by an earlier #19117, we probably had that part of CI not working back then.