AudioResponse.audio now accepts string (base64) or { b64Json, contentType }
object form. Adds "audio-gen" and "fal-audio" endpoint types to FixtureMatch,
RecordProviderKey, and router compatibility checks. Exports FORMAT_TO_CONTENT_TYPE
from helpers. Guards speech handler against object-form audio.

Handles AudioResponse fixtures in both non-streaming and streaming
Gemini paths by emitting inlineData parts with base64 audio. Adds
reasoning/thought accumulation and audio inlineData extraction to
collapseGeminiSSE. Fixes Cohere SSE collapse to preserve content
alongside tool calls.

Handles AudioResponse fixtures by sending inlineData frames with
turnComplete. Adds ContentWithToolCallsResponse handler that streams
text then sends tool calls. Filters thought parts from text extraction.
Pre-computes tool call IDs for wire/history consistency.

Handles /v1/sound-generation and /v1/music/* endpoints. Supports
sound-generation (text field), music compose/stream/plan (prompt field
with composition_plan fallback). Returns binary audio, chunked stream,
or JSON plan based on subType. Full journal integration with fixture
matching and proxy-and-record fallback.

Handles /fal/queue/submit, /fal/queue/requests (status/result/cancel),
and /fal/run endpoints. FalJobMap singleton with TTL and size-bound
eviction. Translates AudioResponse to fal.ai file objects with computed
sizes. Full journal integration, proxy-and-record fallback, test-ID
scoped job isolation.

…thods

Adds server routes for ElevenLabs (/v1/sound-generation, /v1/music/*),
fal.ai (/fal/queue/*, /fal/run/*) with CORS, chaos, and journal.
Adds falJobs.clear() to reset handler and LLMock.reset(). Adds
onAudio, onSoundEffect, onMusic, onFalAudio convenience methods.
Fixes nextRequestError splice race.

Detects audio inlineData in both non-streaming JSON and streaming SSE
Gemini responses, producing AudioResponse fixtures with b64Json and
contentType. Widens recorder EndpointType to include audio-gen and
fal-audio. Adds Float32Array alignment guard for embeddings.

6 new/modified test files covering ElevenLabs sound/music, fal.ai
queue lifecycle, Gemini HTTP audio (streaming + non-streaming),
Gemini WebSocket audio, audio recording/replay, and multimedia
type guard + endpoint filtering.