detect all gpus during hw check

[JannikSt]

A machine might have multiple GPUs installed - previously it just used the first GPU and simply increased the count.

[Copilot]

Pull Request Overview

This PR updates the hardware check logic to detect all GPUs instead of only the first one, enabling better support for multi-GPU systems.

• In worker/src/checks/hardware/hardware_check.rs, the function now collects all GPUs and selects the primary GPU based on a max_by_key filter.
• In worker/src/checks/hardware/gpu.rs, the GPU detection logic has been refactored to return a vector of GPUs and merges devices with the same name.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
worker/src/checks/hardware/hardware_check.rs	Updated to collect multiple GPU specs and select a main GPU from the list.
worker/src/checks/hardware/gpu.rs	Refactored GPU detection to gather all devices, convert from GB to MB, and aggregate same-named GPUs.

Comments suppressed due to low confidence (2)

worker/src/checks/hardware/hardware_check.rs:148

• [nitpick] Selecting the main GPU solely based on 'count' may not accurately represent the best GPU in terms of performance. Consider evaluating other attributes, such as memory_mb, to determine the primary GPU.

.max_by_key(|gpu| gpu.count.unwrap_or(0));

worker/src/checks/hardware/gpu.rs:91

• [nitpick] Merging GPU devices based solely on name may inadvertently combine devices with different specs. Consider verifying that merging is appropriate or handling discrepancies in attributes like memory and driver_version.

existing_device.count += 1;

[mattdf]

Would be useful to change this data structure to contain the cuda indexes of these GPUs, so that when you create a docker container instead of passing --gpus all you could have the option to do docker run --gpus '"device=3,4,5"', and as such removes the need to worry about handling random indexes for mixed card types from inside the worker containers.