docs: add Gradium custom voice and custom transcriber guides

[timpratim]

Description

Adds integration guides for Gradium, a real-time audio
model provider, covering both speech-to-text and text-to-speech. Both connect
through Vapi's existing custom endpoints, so no API or spec changes are needed.

• New page: customization/custom-voices/gradium.mdx — connect the Gradium
  Text-to-Speech API through a custom-voice endpoint. Covers the
  voice-request payload, streaming raw PCM back, sample rates, pooled
  multiplexed connections, pronunciation dictionaries, and a
  fallback voice.
• New page: customization/custom-transcriber/gradium.mdx — connect the
  Gradium Speech-to-Text API through a custom-transcriber WebSocket bridge.
  Covers the start frame, forwarding only channel 0 so the agent transcribes
  the caller alone, the flush handshake that keeps the end of an utterance
  intact, and using Gradium's semantic VAD as the end-of-turn authority via
  smartEndpointingPlan.provider: "custom-endpointing-model".
• docs.yml — adds Gradium under the existing Custom voices section, and
  converts the Custom transcriber page into a section with a Gradium child,
  mirroring how Custom voices is structured.
• Both pages end with a troubleshooting table of the failure modes this
  integration actually hits (self-transcription, truncated finals, endpointing
  races, sample-rate mismatch).
• Code samples are Python, matching Gradium's SDK, and are drawn from a working
  Vapi + Gradium voice agent rather than written from the API reference.

Testing Steps

• Run the app locally using fern docs dev or navigate to the preview deployment
• Ensure that the changed pages and code snippets work