[UR][Offload] Tracking issue for features missing in the offload UR adapter

[RossBrunton]

This is an incomplete list of features that the Offload adapter needs to implement:

Entry Points:

• [optional] All bindless image entry points
• [optional] All command buffer entry points
• [optional] All "physical mem" entry points
• [optional] All "virtual mem" entry points
• [optional] Get/Create with native handle for all modules
• [optional] urAdapterGetLastError
• [optional] urContextSetExtendedDeleter
• [optional] urDeviceGetGlobalTimestamps
• urEnqueueDeviceGlobalVariableRead
• urEnqueueDeviceGlobalVariableWrite
• urEnqueueEventsWait
• urEnqueueEventsWaitWithBarrier
• urEnqueueMemBufferCopy
• urEnqueueMemBufferCopyRect
• urEnqueueMemBufferFill
• urEnqueueMemBufferReadRect
• urEnqueueMemBufferWriteRect
• urEnqueueMemImageCopy
• urEnqueueMemImageRead
• urEnqueueMemImageWrite
• [optional] urEnqueueTimestampRecordingExp
• [optional] urEnqueueNativeCommandExp
• urEnqueueUSMFill
• urEnqueueUSMAdvise
• urEnqueueUSMMemcpy
• urEnqueueUSMPrefetch
• urEnqueueReadHostPipe
• urEnqueueWriteHostPipe
• [optional] urEnqueueUSMFill2D
• [optional] urEnqueueUSMMemcpy2D
• [optional] urEventGetProfilingInfo
• urEventSetCallback
• [optional] urKernelGetSubGroupInfo
• urKernelSetArgLocal
• urKernelSetArgMemObj
• urKernelSetArgSampler
• [optional] urKernelSetSpecializationConstants
• [optional] urKernelGetSuggestedLocalWorkSize
• urMemBufferPartition
• urMemImageCreate
• urMemImageGetInfo
• urQueueFlush
• [optional] urProgramBuildExp
• [optional] urProgramCompileExp
• urProgramGetBuildInfo
• urProgramGetGlobalVariablePointer
• urProgramLink
• [optional] urProgramLinkExp
• [optional] urProgramSetSpecializationConstants
• urSamplerCreate
• urSamplerGetInfo
• urSamplerRelease
• urSamplerRetain
• urUSMGetMemAllocInfo
• [optional] urUSMPitchedAllocExp
• urUSMPoolCreate
• urUSMPoolRetain
• urUSMPoolRelease
• urUSMPoolGetInfo

Info Queries:

• Adapter
• Context
• Device
• Event
  • UR_EVENT_INFO_COMMAND_EXECUTION_STATUS
• Kernel
• Platform
• Program
• Queue
  • UR_QUEUE_INFO_CONTEXT
  • UR_QUEUE_INFO_DEVICE
  • UR_QUEUE_INFO_DEVICE_DEFAULT
  • UR_QUEUE_INFO_SIZE
  • UR_QUEUE_INFO_EMPTY

Queue Flags:

• UR_QUEUE_FLAG_DISCARD_EVENTS
• UR_QUEUE_FLAG_LOW_POWER_EVENTS_SUPPORT_EXP
• UR_QUEUE_FLAG_ON_DEVICE
• UR_QUEUE_FLAG_ON_DEVICE_DEFAULT
• UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE
• UR_QUEUE_FLAG_PRIORITY_HIGH
• UR_QUEUE_FLAG_PRIORITY_LOW
• UR_QUEUE_FLAG_PROFILING_ENABLE
• UR_QUEUE_FLAG_SUBMISSION_BATCHED
• UR_QUEUE_FLAG_SUBMISSION_IMMEDIATE
• UR_QUEUE_FLAG_SYNC_WITH_DEFAULT_STREAM
• UR_QUEUE_FLAG_USE_DEFAULT_STREAM

Misc:

• getInfo doesn't seem to verify the property size in all cases
• workDims > 1 support for urEnqueueKernelLaunch
• UR_DEVICE_INFO_USM_* should return correct capability flags
• Ability to report support for different floating point types, sycl currently expects UR_DEVICE_INFO_*_VECTOR_WIDTH queries (currently implemented with placeholder values) and an extension string for this (pending [UR] Stop querying adapter fp16/fp64 support via extension. #15811)