Moving GPU memory management to OperationManager (clean)

[BenYang2002]

Supersedes

This PR supersedes #841.

The previous pull request contained too many auto-generated or formatting-related file changes, which made it difficult to review effectively. This PR presents a cleaner version of the same core changes, with unnecessary noise removed, and incorporates the feedback provided in #841.

Related Issues

Closes #131
Closes #821
Closes #825

Description

This PR completes the integration of GPU memory management operations — copyFromGPU and deallocateGPUMemory — into the OperationManager framework, extending the pattern previously established with copyToGPU. All GPU operations are now standardized as std::function<void()> callbacks registered during class construction. This change eliminates hardcoded call chains in GPUModel, centralizes GPU logic dispatch, and improves modularity and maintainability.

Key changes include:

• Refactored copyFromGPU and deallocateGPUMemory to take no parameters
• Registered operations using OperationManager::execute(...)
• Moved operational context into class member variables
• Added #ifdef USE_GPU guards where appropriate
• Preserved consistency with the previously refactored copyToGPU logic

[stiber]

Not sure if this is ready for review, as no request was made. I will point out some things:

• Make sure that GPU regression tests have been run manually and note that in the conversation.
• This PR can't be simply merged. Checking the different branches, it appears that it is three commits behind SharedDevelopment. Follow our SWE process and merge SharedDevelopment into this branch, re-test it and ensure it works, then request again that it be reviewed.

[NicolasJPosey]

@NicolasJPosey to run GPU regression and report back on pass/fail status.

[stiber]

Pending GPU testing and approval by @NicolasJPosey, this looks good to go!

[NicolasJPosey]

@stiber I am not sure how it got through, but a device side function that was previously in GPUModel.cpp was causing a compile error. This was moved into the AllSpikingNeurons_d.cpp a while back so I just removed it from GPUModel. The medium connected dataset passed regression testing and the large long dataset ran to completion without any errors so I will approve the PR. Since I committed a new change, you may have to re-approve.

[NicolasJPosey]

GPU regression test-medium-connected and test-large-long both pass.