Make pre-detection audio clip offset user-configurable

[tphakala]

Currently, BirdNET-Go uses a fixed pre-detection offset for starting audio clip capture. On my systems, I've noticed that sometimes the audio capture starts too late, potentially cutting off part of the bird call. The reasons for this could be numerous.

In any case, I think it would be beneficial to make the pre-detection recording offset user-configurable within a reasonable limit, such as allow capture up to 10 seconds of audio before event.

This likely requires extending total capture length by same amount. At the moment there is a depencendy of fixed capture length on spectrogram generation which should be addressed.