Remove `check_requirements(('tensorflow>=2.4.1',))` (#5476)

`check_requirements()` is unreliable for large packages like torch and tensorflow that may have multiple installation routes (i.e. conda, pip, tensorflow-cpu, etc.)

detect.py

@@ -94,7 +94,6 @@ def run(weights=ROOT / 'yolov5s.pt',  # model.pt path(s)
             import onnxruntime
             session = onnxruntime.InferenceSession(w, None)
     else:  # TensorFlow models
-        check_requirements(('tensorflow>=2.4.1',))
         import tensorflow as tf
         if pb:  # https://www.tensorflow.org/guide/migrate#a_graphpb_or_graphpbtxt
             def wrap_frozen_graph(gd, inputs, outputs):