fix libcuda path when running in WSL (chapel-lang#24347)

This PR adds the ability to detect if we are using an NVidia GPU while
running in Windows Subsystem for Linux (WSL) and make appropriate
additions to properly detect the location of `libcuda`.

This change was needed because WSL has a special configuration w.r.t.
the GPU drivers where the drivers are installed only in Windows and then
made available to WSL.

TESTING:

- [x] paratest on chapdl-nvidia `[Summary: #Successes = 176 | #Failures
= 0 | #Futures = 6]`
- [x] code targeting GPU compiles and runs in WSL with an NVidia GPU

[reviewed by @e-kayrakli - thanks!]

util/chplenv/chpl_platform.py

@@ -76,6 +76,12 @@ def get(flag='host'):
     return platform_val
 
 
+@memoize
+def is_wsl():
+    name = (platform.uname().release).lower()
+    if name.endswith('-microsoft') or name.endswith('-microsoft-standard-wsl2'):
+        return True
+
 @memoize
 def get_mac_os_version():
     release, version, machine = platform.mac_ver()

util/chplenv/compile_link_args_utils.py

@@ -100,8 +100,11 @@ def get_runtime_link_args(runtime_subdir):
         sdk_path = chpl_gpu.get_sdk_path(gpu_type)
         if gpu_type == "nvidia":
             system.append("-L" + os.path.join(sdk_path, "lib64"))
-            system.append("-lcuda")
             system.append("-lcudart")
+            if chpl_platform.is_wsl():
+                # WSL needs to link with libcuda that belongs to the driver hosted in Windows
+                system.append("-L" + os.path.join("/usr", "lib", "wsl", "lib"))
+            system.append("-lcuda")
         elif gpu_type == "amd":
             lib_path = os.path.join(sdk_path, "lib")
             system.append("-L" + lib_path)