Split VkCmd recorded commands into their own category

[olehkuznetsov]

Description

This PR splits vkCmd* recorded commands into their own CPU timing category (VulkanCPUTiming/VkCmd), separating them from other VkCommandBuffer operations (such as vkBeginCommandBuffer and vkEndCommandBuffer).

Motivation

Previously, all VkCommandBuffer related calls were grouped under a single Perfetto category (VulkanCPUTiming/VkCommandBuffer). Enabling this category meant tracing every single recorded command (e.g., vkCmdDraw, vkCmdBindPipeline), which introduces massive CPU overhead and drastically inflates trace sizes due to their very high frequency during frame recording.
By splitting recorded commands into a distinct VkCmd category, users can now selectively trace important command buffer lifecycle operations (like Begin and End) using the existing VkCommandBuffer category. This allows for logging important buffer commands without incurring the huge performance penalty of tracing every individual vkCmd* call, unless explicitly needed.