Testing: add unit tests for gpu_detection.py

[mrveiss]

Problem

autobot-backend/utils/gpu_optimization/gpu_detection.py has zero test coverage. Recent refactors (#1990, #2222) introduced lru_cache-based vendor caching and a module-level GPU name stash that especially need test coverage.

Discovered During

Code review of PR #2237 (fix for #2222).

Tests Needed

• _check_nvidia_gpu() returns GPU name string (not bool) or None
• _check_nvidia_gpu() takes only first line on multi-GPU output
• _detect_vendor() caches result and stashes NVIDIA GPU name
• _detect_vendor() returns correct vendor for AMD and Intel
• _detect_vendor() returns None when no GPU present
• _reset_detection_state() clears both lru_cache and _nvidia_gpu_name
• _detect_nvidia_capabilities() uses stashed name and parses two-field output
• _detect_nvidia_capabilities() takes first line on multi-GPU output
• check_gpu_availability() and detect_gpu_capabilities() use cached vendor
• Cache clear + re-detection produces correct state (no stale data)

All subprocess calls should be mocked (unittest.mock.patch).

Impact

Medium — no production risk, but the caching/stashing logic is easy to break without test coverage.