This test exercises `bst artifact checkout --pull <artifact name>`

New regression test to ensure we continue to support `bst artifact pull` properly for artifacts which are not locally cached.

Instead of calling self._scheduler.clear_queues(), add a focus point
for resetting state related to a scheduler run. I suspect that the
session element counts have been flawed in cases where we run the
scheduler multiple times but fail to clear that part of the state.

This should all around be a better approach for multiple scheduler runs.

In any case that the user has requested artifacts be pulled, and that
artifact targets have been specified, pre-emptively pull the target
artifact metadata so that the operation can complete coherently.

Specifically, this fixes:

  * bst artifact pull --deps build <artifact name>
  * bst artifact checkout --deps build --pull <artifact name>

In these cases, we cannot know the build dependencies of the artifact
if we do not at least try to obtain the toplevel target artfiact(s) first,
so we instate a policy to initially download metadata.

This introduces a new parameter to the general purpose _load() function,
which is called `attempt_artifact_metadata`.

When `attempt_artifact_metadata` is specified, we will attempt to download
artifact metadata for toplevel specified targets, this way we can derive
their dependencies which is needed for some operations.

Currently this unfortunately downloads the entire artifact, but should
be optimized later on to download only the required metadata.

When an element finishes being pulled, call _mimic_artifact() so
that the element assumes the properties of the downloaded artifact.

This function causes the element to mimic it's artifact instance,
and is initially used when instantiating from an artifact.

This needs to be exposed so that ArtifactElement can trigger this
activity in response to _pull_done().

Additionally, this adds the declaration of __environment and
__variables to the __init__ function, as I received some complaints
from pylint and mypy from this commit.

This is merely because the structure of remotes is heavily bound to
loaded projects, and in some cases we need to download some artifacts
before some ArtifactProjects can exist and then later pull elements
related to those newly created ArtifactProject instances.

This is rather a silly workaround, instead the data structure of
remotes should be reorganized, which is a larger task and probably appropriate
to examine when re-examining remote service configuration surfaces.

_stream.py, _frontend/widget.py: Fix weird hack

See merge request BuildStream/buildstream!2117

When stream is asked for a list of artifacts to show for
the purpose of `bst artifact show`, it was squashing the element
name with the artifact name before it gets displayed in the
frontend.

Instead, make the special casing in the frontend.

Require all stack dependencies to be both build & runtime dependencies

Closes #1075

See merge request BuildStream/buildstream!2113