Fix nvexec bulk callable forwarding

[fallintoplace]

What changed

• Preserve the callable's cvref when connecting the single-GPU nvexec bulk sender.
• Add coverage for running a const lvalue bulk sender.

Why

bulk_sender::connect() always cast the stored callable to Fun&&. That can move from an lvalue sender and makes const lvalue senders fail to compile. The sender's value category should decide whether the callable is moved or copied.

Testing

• git diff --check
• CUDA regression test added for GPU CI

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[ericniebler]

/ok to test ba3fda5

[ericniebler]

/ok to test e0c6df2