cuda: Allow for and restrict use of Cuda Toolkits to >= 11.4.

[Treece-Burgess]

Pull Request Description

In the master branch, the cuda component supports down to Cuda Toolkit 11.5; however, support can go down even further to Cuda Toolkit 11.3 when the MetricsEvaluator API was initially introduced.

This PR adds support for Cuda Toolkit 11.4 and also restricts users at runtime to need to use Cuda Toolkits >= 11.4. The restriction is added as Cuda Toolkit 11.3 suffers from the same issue as discussed on the following NVIDIA Developers Forum post. From my testing 11.4 does not suffer from this.

Testing

Setup

Testing was done on Athena at Oregon, with the following setup:

• OS: RHEL 8.9
• CPU: AMD EPYC 7662
• GPUs: 4 * A100
• Cuda Toolkits 11.3, 11.4, and 12.8.1

Results

• PAPI compilation: ✅ (with Cuda Toolkits 11.3, 11.4, and 12.8.1)
• PAPI utilities*: ✅

Using Cuda Toolkit 11.3, shows that the cuda component is disabled
Name:   cuda                    CUDA profiling via NVIDIA CuPTI interfaces
   \-> Disabled: The cuda component supports Cuda Toolkits >= 11.4.

• Cuda component tests: ✅ (with Cuda Toolkits 11.4 and 12.8.1)

* papi_component_avail, papi_native_avail, and papi_command_line

Author Checklist

• Description
  Why this PR exists. Reference all relevant information, including background, issues, test failures, etc
• Commits
  Commits are self contained and only do one thing
  Commits have a header of the form: module: short description
  Commits have a body (whenever relevant) containing a detailed description of the addressed problem and its solution
• Tests
  The PR needs to pass all the tests