Add MiniMax text to speech support to the speech generation path

[octo-patch]

Reason: add MiniMax text to speech support to the existing speech generation path

Summary

• Route the existing speech generation helper to the MiniMax text to audio operation when
  jeecg.ai-chat.ai-model-voice.provider is MINIMAX, keeping the current OpenAI compatible
  /audio/speech branch as the default so the voice module and the video dubbing caller are unchanged.
• Add MiniMaxVoiceApi with the regional /v1/t2a_v2 endpoints, the accepted speech models and the
  default model, the documented request fields (model, text, stream, output_format,
  voice_setting, audio_setting) and Bearer authorization.
• Parse the json response instead of streaming raw bytes: validate base_resp.status_code, then decode
  the hex encoded data.audio payload and write it to the target file, reporting data.status when no
  audio is returned.
• Accept every configured api host shape (bare regional host, host carrying the version segment, or the
  full operation url) so the endpoint is taken from configuration rather than assumed, and clamp the
  configured speed and decibel volume gain into the ranges the operation accepts.

Changes

• jeecg-boot/jeecg-boot-module/jeecg-boot-module-airag/src/main/java/org/jeecg/modules/airag/voice/util/MiniMaxVoiceApi.java (new)
• jeecg-boot/jeecg-boot-module/jeecg-boot-module-airag/src/main/java/org/jeecg/modules/airag/voice/util/VoiceApiHelper.java
• jeecg-boot/jeecg-boot-module/jeecg-boot-module-airag/src/test/java/org/jeecg/modules/airag/voice/util/MiniMaxVoiceApiTest.java (new)

Verification

• javac compile of the changed speech sources together with AiChatConfig on Java 21: passed.
• MiniMaxVoiceApiTest on the JUnit 5 console launcher: 16 tests passed. The suite drives
  VoiceApiHelper.generateAudio end to end against a local http server and asserts the request path,
  the Bearer header, the request body fields, and that the decoded audio bytes reach the output file.
  It also covers a failed base_resp.status_code, a non 200 response, malformed hex payloads, and a
  regression test that a non MiniMax provider still calls /audio/speech.
• Maven could not resolve the full reactor in this environment, so the module build was not run and is
  not claimed here.