Feature request: add Gemma 4 audio input support

[scouzi1966]

Summary\nAdd first-class audio input support for Gemma 4 models in AFM.\n\n## Motivation\nIssue #82 focused on Gemma 4 tool-calling and reasoning behavior. Audio input support was identified as a separate, larger piece of work and should be tracked independently as a feature request.\n\n## Goal\nSupport audio inputs for Gemma 4 requests through the AFM API when the underlying model/runtime path supports them.\n\n## Proposed scope\n- confirm whether the current MLX / vendor Gemma 4 path supports audio as a first-class input\n- if supported, extend request decoding to accept audio content parts\n- route audio media through MLXModelService extraction / preprocessing\n- expose audio capability metadata where appropriate\n- add unit tests for request decoding, routing, and unsupported-model rejection\n\n## Notes\nThis issue is intentionally separate from #82 so tool-calling and parser work can be closed independently of audio support.\n\nRelated: #82