Qualcomm AI Engine Direct - Add 4-bit Embedding Quantization Option (#7691)

* Qualcomm AI Engine Direct - Add 4-bit Embedding Quantization Option

Summary:
 - Introduce 4-bit embedding quantization for prefill, kv, and hybrid mode
 - Fixe an assertion condition bug in the annotate_and_quant_scalar pass
 - Refactor passes in capture_program
 - Add topological sorting for passes in capture_program

* fix wrong import