Fixes OOM Errors - too high RAM usage by VAD

Reported problems:
#1193
#1169

VAD implementations consumes humongous memory amounts [original Silero doesn't have this problem]

This PR should fix the OOM problem.
Alt solution could be removing 'lru_cache'.

faster_whisper/vad.py

@@ -260,8 +260,9 @@ def __init__(self, encoder_path, decoder_path):
             ) from e
 
         opts = onnxruntime.SessionOptions()
-        opts.inter_op_num_threads = 0
-        opts.intra_op_num_threads = 0
+        opts.inter_op_num_threads = 1
+        opts.intra_op_num_threads = 1
+        opts.enable_cpu_mem_arena = False
         opts.log_severity_level = 4
 
         self.encoder_session = onnxruntime.InferenceSession(