Set train encoder to True for skill chain config

[gchhablani]

We need to set train_encoder: True otherwise the high-level policy tries to load a checkpoint into *.visual_encoder which does not exist.

[kason-huang]

After I set the train_encoder: True, and I run the command

python -um goat_bench.run   --run-type train \
  --exp-config config/experiments/ver_goat_skill_chain.yaml \
  habitat_baselines.num_environments=4 \
  habitat_baselines.tensorboard_dir=${TENSORBOARD_DIR_} \
  habitat_baselines.checkpoint_folder=${SKILL_CHAIN_CHECKPOINT_DIR}

But I but got the following response