amdgpu: fix AMDGPU stats showing 0 due to invalid gpu_metrics data

[lullptr]

Issue: AMDGPU stats displaying 0
Fixes #2028
Fixes #2038

Problem

I understand that the current implementation uses a hybrid model where sysfs values act as the baseline and are overridden by gpu_metrics when available.

However, the override logic assumes that the presence of the gpu_metrics file implies valid and supported data. In practice, this is not always true, the file may exist and be readable even when the kernel/firmware reports unsupported or invalid metrics.

As a result, valid sysfs values are unconditionally overwritten by invalid gpu_metrics data, leading to GPU statistics incorrectly showing as 0.

Fix

Introduce a validity tracking flag for gpu_metrics sampling.

The flag is set only when at least one gpu_metrics sample successfully produces valid and usable metric data. This indicates gpu_metrics is functional on the system and only then is gpu_metrics allowed to override sysfs values.

Consideration

This fix preserves the existing hybrid model by allowing gpu_metrics to override sysfs values once at least one valid sample has been observed.

However, it may be worth reconsidering whether a single successful sample is sufficient, or if multiple consecutive valid samples should be required before enabling overrides.

Screenshots

Before patch

After patch