Fallback to CPU if preconfigured inference device is unavailable

[leoll2]

Issue #5042 enables users to configure what device to use for inference (CPU, GPUs, ...). The selected device is saved in the pipeline entity, so that the choice remains in effect even if the user restarts the application, or re-activates the pipeline after temporarily switching to another one.

There could be edge cases where the previously configured device is actually not available at runtime, e.g. if the application is restarted after an hardware reconfiguration. To withstand such scenarios, we need a mechanism that fallbacks to CPU when the configured inference device is not accessible.

Task

• When loading the inference pipeline, check if the configured device is present in the list of available devices. If not, log a warning, reconfigure to CPU and retry the load.