Add array-based acoustic APIs: predict_arrays() and encode_arrays()

[fegue]

Summary

Follow up on #47 by adding array-based acoustic APIs to birdnetR.

The file-based encoding API was added in #47, but the array-based APIs were intentionally deferred to keep scope narrow.

Motivation

The underlying Python birdnet package exposes both:

• predict_arrays()
• encode_arrays()

In birdnetR, array-based support should be added as a coherent pair rather than exposing only one of them. This keeps the acoustic API symmetric across input modalities:

• file input: predict(), encode()
• array input: predict_arrays(), encode_arrays()

Proposed scope

• add predict_arrays() for birdnet_model_acoustic
• add encode_arrays() for birdnet_model_acoustic
• define and document the R input contract for in-memory audio arrays and sample rates
• add tests for R-to-Python conversion and result handling
• update documentation and pkgdown reference

Notes

Potential design questions to resolve:

• preferred R input structure for audio arrays plus sample rate
• validation rules for array shape and type
• whether helper utilities are needed for user-friendly construction of array inputs