Increase CPU subprocess timeout from 10min to 2h for long recordings

Usage logging V2: grouped JSON schema, endpoint tracking, GPU profiling

Remove dead code and add GPU VRAM profiling

Isolate CPU fallback in spawn subprocess to prevent CUDA state poisoning

Eliminate in-decorator CPU fallback to prevent CUDA state poisoning

Stop CUDA state poisoning by not invalidating model caches during quota fallback

Add debug logging

Park AoTI cache and detach VAD tensors to prevent CUDA state poisoning

Fix syntax

Prevent CPU fallback from poisoning CUDA state via GPU model parking

Limit CPU fallback to quota errors only

Refine ZeroGPU fallback for worker/runtime errors

Reset torch.cuda._initialized on GPU errors instead of nuking global model caches

Replace os._exit(1) with CPU fallback on GPU errors

Add transition segment detection for non-Quranic phrases (Amin, Takbir, Tahmeed)

Add GPU retry for SDK worker errors instead of immediate CPU fallback

Fix SDK worker_init CUDA poisoning: reset immediately instead of 300s cooldown

Fix GPU cross-user contamination: scope errors per-request, not process-wide

Fix GPU quota flag blocking unrelated users across the shared process

Eliminate CUDA state poisoning at root cause, not just detection

Catch SDK-wrapped GPU errors that bypass CUDA error detection

Fix CUDA state poisoning by eliminating CUDA ops outside GPU leases

Prevent CUDA state poisoning and add thread-safe GPU lifecycle management

Harden GPU fallback to prevent CUDA init from permanently poisoning process

Compress logging audio format

Fix race condition where concurrent quota exhaustion permanently breaks GPU

Widen DP lookback window and fix null-typed parquet columns

Fix delay between progress bar completion and results displaying

Restructure src/ for cleaner modularity