support flexible learning rate adjustment strategy

[jacquesqiao]

• learning rate should be a global variable. depends on Need a global counter for training. #7784
• define the interface to do learning rate decay. pd-refs tf-refs:
  • framework Add learning rate decay #7892
  • exponential_decay Add learning rate decay #7892
  • inverse_time_decay Add learning rate decay #7892
  • piecewise_decay Add polynomial_decay and piecewise_decay #8013
  • polynomial_decay Add polynomial_decay and piecewise_decay #8013
• use condition op to adjust the learning rate during training. Add polynomial_decay and piecewise_decay #8013 Impl scalar switch case op with condition op #8184
• add elementwise_pow_op [wip]Add elementwise pow op #7911
• support syntax like a+b, a*10. Add math operator patches #7688
• fix floor op fix floor_op #7926
• what device should learing rate decay happen? should it all happen on CPU? Use force cpu in fill constant op #8254
• Add switch op Impl scalar switch case op with condition op #8184

[reyoung]

This issue depends on #7784

[jacquesqiao]

all task is done now