Refactor transcription flow to use single backend job per audio file

Summary

This MR refactors the normal backend transcription flow to process the entire uploaded audio file using a single backend transcription job instead of chunk-based uploads.

Previously, the frontend decoded and split audio files into chunks client-side, where each chunk independently triggered api.transcribeAudio(...), resulting in multiple /api/transcribe requests, multiple job_ids, and chunk-wise transcription output.

With this update, the frontend now uploads the complete audio file once using api.transcribeFile(...) and polls a single job_id until processing is completed.

Changes Made

• Removed chunk-based transcription flow for normal backend mode.
• Updated frontend to upload the entire audio file in a single request.
• Refactored polling logic to handle only one transcription job.
• Updated UI state handling to display transcription output only after full processing completion.
• Preserved existing chunking behavior for diarization mode where required.

Benefits

• Aligns frontend behavior with backend processing flow.
• Reduces unnecessary API calls and polling requests.
• Improves transcription consistency and response handling.
• Prevents partial/chunk-wise transcription rendering.
• Simplifies job tracking and state management.

Affected Areas

• Audio transcription flow
• API handling (api.ts)
• Job polling mechanism
• Transcription result rendering

Testing Performed

• Verified single /api/transcribe request is created per uploaded audio file.
• Confirmed only one job_id is generated and polled.
• Validated complete transcription output is returned as a single response.
• Ensured diarization mode functionality remains unaffected.