Init/apply is broken for remote backends

Original init command works if we add TF_WORKSPACE env var, but apply for some reason runs init again, but explicitly unsets TF_WORKSPACE, resulting in a error:

export workspace
unset TF_WORKSPACE
...
terraform init -input=false -no-color $INIT_ARGS "$module_path"
terraform workspace select -no-color "$workspace" "$module_path" || terraform workspace new -no-color "$workspace" "$module_path"
terraform apply -auto-approve $PLAN_ARGS "$module_path"
Initializing modules...

Initializing the backend...

The currently selected workspace (default) does not exist.
  This is expected behavior when the selected workspace did not have an
  existing non-empty state. Please enter a number to select a workspace:
  
  1. test-ci

  Enter a value: 

[whyman]

Hi @brunomed,

The orb is designed to take workspace as a parameter rather than have it set via an env var.

Does it work as expected using this method?