Changed "face_recognition_demo" - fixing the device argument to accept AUTO: GPU,CPU style inputs for flexibility

[InfoSage05]

Summary

• I have removed choices=DEVICE_KINDS restriction from -d_fd, -d_lm, and -d_reid
  arguments in face_recognition_demo.py
• The old code only accepted bare device names (CPU, GPU, AUTO, HETERO) and
  rejected valid OpenVINO compound strings like AUTO:GPU,CPU or HETERO:GPU,CPU
  with an argparse error so I have added them in the argument.
• I have also updated help text for all three arguments to document the correct compound format

Root Cause

OpenVINO supports compound device strings as a standard way to configure
the AUTO and HETERO plugins. For example:

• AUTO:GPU,CPU — automatically selects the best device, and it prefers GPU
  and falls back to CPU
• HETERO:GPU,CPU — splits model layers across GPU and CPU

These strings are not standalone device names and they carry a priority list
after the colon. However, the demo used Python's argparse with
choices=DEVICE_KINDS where:

 DEVICE_KINDS = ['CPU', 'GPU', 'NPU', 'AUTO', 'HETERO'] 

 argparse choices= performs exact string matching against the list.
 So when a user passed -d_fd AUTO:GPU,CPU, argparse compared AUTO:GPU,CPU
 against each entry in the list, found no exact match, and immediately raised
 an error:

 error: argument -d_fd: invalid choice: 'AUTO:GPU,CPU'
 (choose from CPU, GPU, NPU, AUTO, HETERO)

 This means the AUTO plugin could never be used to its full potential from
 the command line  and users were silently limited to just the AUTO feature with no device
 priority control.

 Solution

 Removed the choices=DEVICE_KINDS restriction from the three device
 arguments -d_fd, -d_lm, and -d_reid. The arguments now accept any
 string, which allows all valid OpenVINO device specifiers including compound
 ones. The help text for each argument was also updated to explicitly document
 the AUTO:<dev1>,<dev2> and HETERO:<dev1>,<dev2> formats so users know
 what is supported.  

How to Test

python face_recognition_demo.py -i <input> -m_fd <model> -m_lm <model> \
  -m_reid <model> -fg <gallery> -d_fd AUTO:GPU,CPU

[workflow-lab]

Can one of the admins verify this patch?