Prediction: use acceleration from RNN model and add a gflag in conf file for tim thred to prediction trajectories

* prediction: add rnn network proto

* Prediction: added build support for rnn network proto (#902)

* Prediction: added build support for rnn network proto

* Prediction: fixed an assertion bug

* prediction: add rnn model

* Prediction: Lint fixes

* Prediction: added RNN evaluator framework

* prediction: add rnn model file

* Prediction: build up functional framework for RNN evaluator

* Prediction: implement RNN model extract obstacle features

* Prediction: fixed a few lint errors

* Prediction: implement set lane feature

* Prediction: fixed cut in trajectory when obstacle is close to ego vehicle

* Prediction: init rnn state

* Prediction: a little more

* Prediction: almost implement Evaluate in RNN

* Prediction: implement load model

* Prediction: add conf for RNN model

* prediction: refine rnn model

* prediction: fixbug of rnn state

* prediction: increase minimun prediction length

* Prediction: move a hard-coded parameter to gflags

* Prediction: restructured network and fixed unit test

* prediction: add unit test for rnn_model (#1132)

* prediction: add acceleration in rnn model

* Prediction: add cyclist model

* Prediction: updated predictor manager to read configuration for cyclists (#1197)

* drivers: using cvar to trigger publishment (#931) (#1200)

* Prediction: use kf gflag to trigger motion part in move sequence predictor (#1282)

* prediction: add documents for network

* Prediction: a light weights rnn model

* Prediction: add gflag to use lane sequence acc