Separately version the CUDA compiler

[maleadt]

ptxas and friends are now (unconditionally) provided by CUDA_Compiler_jll, sourced from either CUDA 11.8 or 12.9 depending on which driver is used. This makes sure we're using the latest compiler, which fixes bugs (especially 11.4 and 11.5 had plenty of them) and ensures we can target recent devices and PTX ISAs regardless of the runtime selected by the user.

This should also facilitate using NVVM instead of NVPTX, which doesn't support targeting older PTX ISAs.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic master) to apply these changes.

Click here to view the suggested changes.

diff --git a/src/compatibility.jl b/src/compatibility.jl
index 4ffcbce33..addba40b5 100644
--- a/src/compatibility.jl
+++ b/src/compatibility.jl
@@ -56,8 +56,8 @@ const cuda_cap_db = Dict(
     v"8.7"   => between(v"11.4", highest),
     v"8.9"   => between(v"11.8", highest),
     v"9.0"   => between(v"11.8", highest),
-    v"12.0"  => between(v"12.8", highest),
-    v"12.1"  => between(v"12.9", highest),
+    v"12.0" => between(v"12.8", highest),
+    v"12.1" => between(v"12.9", highest),
 )
 
 function cuda_cap_support(ver::VersionNumber)
@@ -293,7 +293,7 @@ function llvm_compat(version=LLVM.version())
     return (cap=cap_support, ptx=ptx_support)
 end
 
-function cuda_compat(driver=driver_version(), compiler=compiler_version())
+function cuda_compat(driver = driver_version(), compiler = compiler_version())
     # devices have to be supported by both the compiler and the driver
     driver_cap_support = cuda_cap_support(driver)
     compiler_cap_support = cuda_cap_support(compiler)