Trackers revamp

[IlyasMoutawwakil]

• Standardizing the latency trackers and making them only track, the timing is handled by a separate class Timer.
• Making the memory tracker more robust when used with launch, by monitoring the main benchmark process (so no need to aggregate VRAM over multiple processes) and supporting both per process and global VRAM tracking at the same time.

[IlyasMoutawwakil]

@franchukpetro I know you've been using VRAM measurements by setting PROCESS_SPECIFIC_VRAM.
This PR removes that environment variable in favor of reporting both (memory.max_global_vram and memory.max_process_vram), hope this makes VRAM benchmarking easier 🤗

[franchukpetro]

Makes sense, thanks a lot for letting me know, @IlyasMoutawwakil!