[SYCL][Graph] Promote graph spec status to experimental

[reble]

This patch updates the status of the SYCL Graph extension from proposed to experimental (after 72341ee).
Including an additional section on unimplemented features.

See https://github.com/reble/llvm#implementation-status for the status
of our total work.

[sommerlukas]

While revisiting the graph fusion proposal in order to update it to match the merged version of graphs, I noticed a few small issues:

• The diamond dependency example using buffers violates the buffer limitations and would result in an error if actually implemented. The example would either need to call buffer::set_write_back(false) on the buffers or use the new property assume_buffer_outlives_graph.
• The example might also be a good place to demonstrate the use of these properties and their effect.
• Super minor nit: The following code snippet for the depends_on property is missing a { after the namespace declaration:

namespace sycl::ext::oneapi::experimental::property::node
class depends_on {
  public:
    template<typename... NodeTN>
    depends_on(NodeTN... nodes);
};
}

[gmlueck]

Sorry, I just noticed that graph_support_level is in the info namespace. That doesn't make sense. The info namespace is reserved for information descriptors, which this is not. I think graph_suppport_level should instead be in sycl::ext::oneapi::experimental.

[EwanC]

The precedent we were basing a info::device::graph_support_level query returning info::graph_support_level on in the Core SYCL spec was info::device::device_type descriptor returning a info::device_type enum class, or info::device::global_mem_cache_type returning a info::global_mem_cache_type enum class.

Made this suggested change to remove the info namespace from sycl::ext::oneapi::experimental::info::graph_support_level, but curious what's different about graph_support_level from info::device_type and info::global_mem_cache_type that means it's not classed as an information descriptor?

[EwanC]

Discovered that changing experimental::info::graph_support_level to experimental::graph_support_level in our implementation is an ABI breaking change, as a symbol with the old namespace was exported from DPC++ at the end of the last window. The next window for ABI breaking changes is currently April, but hopefully we can get an exception based on the wording

Note: Features clearly marked as experimental are considered as an exception to this guideline.

If we can't get an exception, then depending on the timelines we want this extension to "experimental" by, and how big an issue this namespacing is, we could potentially keep it as info::graph_support_level with a spec issue to update it.

[gmlueck]

The precedent we were basing a info::device::graph_support_level query returning info::graph_support_level on in the Core SYCL spec was info::device::device_type descriptor returning a info::device_type enum class

This is a good observation. You convinced me that it is OK to keep info::graph_support_level. So, either way is OK with me. We seem to have precedents for both styles.

[gmlueck]

This looks good. I leave it up to you to decide if you want to change back to info::graph_support_level.