Show layer specified learning rates with combine_tessdata -l

[nagadomi]

combine_tessdata -l ... command shows the initial value of the learning rate, but not the layer (final)learning rates that are actually used in finetuning (lstmtraining with -continue_from option).
This PR will add layer learning rates to the output of combine_tessdata -l command.

Example output:

% ./combine_tessdata -l ../tessdata_best/eng.traineddata
LSTM: network=[1,36,0,1Ct3,3,16Mp3,3Lfys64Lfx96Lrx96Lfx512O1c1], int_mode=0, recoding=1, iteration=814100, sample_iteration=814136, null_char=110, learning_rate=0.001, momentum=0.5, adam_beta=0.999
Layer Learning Rates: :0(Input)=0.001, :1:0(Convolve)=0.001, :1:1(ConvNL)=0.0005, :2(Maxpool)=0.001, :3:0(Lfys64)=0.0005, :4(Lfx96)=0.0005, :5:0(Lrx96)=0.0005, :6(Lfx512)=0.0005, :7(Output)=0.0005

% ./combine_tessdata -l ../tessdata_best/jpn_vert.traineddata              
LSTM: network=[1,48,0,1Ct3,3,16Mp3,3Lfys64Lfx96Lrx96Lfx512O1c1], int_mode=0, recoding=1, iteration=1874000, sample_iteration=1878622, null_char=414, learning_rate=0.001, momentum=0.5, adam_beta=0.999
Layer Learning Rates: :0(Input)=0.001, :1:0(Convolve)=0.001, :1:1(ConvNL)=0.000125, :2(Maxpool)=0.001, :3:0(Lfys64)=0.000125, :4(Lfx96)=0.000125, :5:0(Lrx96)=0.000125, :6(Lfx512)=0.000125, :7(Output)=0.000125

In this output, we can see that the learning rate of jpn_vert has been decayed to 0.000125.
This information may be useful for debugging when finetuning does not work very well, or for determining the learning rate when doing a full training.