zstdgpu/imp: adds public api for single submission mode.

[pm4rtx]

This PR adds a couple of public APIs to simplify necessary steps to submit decompression pipeline into a single command list that becomes possible if the caller can either specify required frame/block-level constant or is fine to rely on library default estimator:

• zstdgpu_SetupAllStageSubmission enables single submission without specifying frame/block-level constants upfront via zstdgpu_SetupFrameInfoConstants / zstdgpu_SetupBlockInfoConstants (command line --ssm)
• zstdgpu_SubmitAllStagesWithExternalMemory, zstdgpu_GetAllStageGpuMemoryRequirement, zstdgpu_SubmitAllStagesWithInteralMemory

Additionally, this PR

• re-visits internal timestamps handling to measure GPU time
• excludes optional zstdgpu-side memory upload from decompression time measurement
• makes sure it's fine to specify all frame/block-level constants that are larger than discovered during data parsing
• always enable SetStablePowerState
• makes sure decompressed frames aren't written to disk unless requested via --out-frm