getDeviceInfo: different errors for no vs old CUDA driver

[Fedr]

Problem

On a computer without any NVIDIA card/driver, cudaGetDeviceCount returns cudaErrorInsufficientDriver — "CUDA driver version is insufficient for CUDA runtime version". That's the same error reported when a real NVIDIA GPU is present but its driver is merely too old for our CUDA runtime. So the two very different situations produced an identical, misleading message, and the user couldn't tell whether they need to install a driver or update one.

Fix

Query cudaDriverGetVersion() first, before cudaGetDeviceCount(). Unlike cudaGetDeviceCount, it returns cudaSuccess even when no driver is installed — it just reports the version as 0 via its out-parameter. So:

• driverVersion <= 0 ⇒ no CUDA driver installed at all → reported as a distinct error and we stop early.
• After that early return, a failure from cudaGetDeviceCount means the driver is present but unusable (too old, or no device).

The installed CUDA driver version is appended to the device-detection error text, so the message is self-diagnosing (e.g. the user can see their driver only supports CUDA 11.4 while the build needs 12.x).

Resulting getDeviceInfo() errors

Situation	Message
No CUDA driver installed	NVIDIA GPU error: no CUDA driver found
Driver present but too old	NVIDIA GPU error: CUDA driver version is insufficient for CUDA runtime version, CUDA driver 11.4
Driver OK, but no GPU	NVIDIA GPU error: no capable device found, CUDA driver 12.2

No functional change for machines with a working CUDA GPU.