Fix: handle _action_masks buffer in single-task scenarios (#67)

update readme + dockerfile

only instantiate termination pred head if episodic=true

only instantiate termination pred head if episodic=true

update readme + clean up

full support for episodic rl

clean up

experimental changes to termination prediction

add walker2d

QoL improvements to termination signal debugging

merge main into branch + fix termination in td-targets

fix multitask model api conversion

update torchrl version

auto-convert old checkpoints to new format

partial fix to loading checkpoints