…eFlow state

[no changelog]

It is being moved into a custom handler for the debug app.

otherwise if _handle_single_message raises an exception (which is
fortunately not something that should generally happen), the
finally-block would fail on "referenced before assignment"

* RequestPaint message is sent at construction time to force calculation
  of number of pages
* given that Attach corresponds to "start the layout" message, Child now
  responds to Attach the same way it responds to RequestPaint, by
  force-repainting everything.

This doesn't actually advance the progress while device erase is ongoing, but at least the user sees a loader.

to prevent "unexpected response to Initialize" errors

The global layout related changes were wrong for T1 where debuglink
behavior is significantly different; in particular, it is not always
possible to communicate over debuglink.

This change reverts to the old behavior for T1B1 and keeps the new one
only for core-based models.

This was necessary for hooking display.refresh() with the old UI toolkit.
With the new one, we explicitly refresh the display after every paint, so
implicit after-step refresh seems no longer necessary.

same behavior as on Trezor One

* improve sync_responses to work on uninitialized instance
* sync responses at construction time

Otherwise a test can hang if the result is not returned. This is not
even covered by pytest-timeout because if a test fails, the call to
task.kill() happens during teardown, and pytest-timeout doesn't cover
that.

In certain cases a DebugLinkState response can get stalled and debuglink
will not read it -- i.e., a situation analogous to wirelink de-sync.
There doesn't seem to be a good way to force-sync debuglink the same way
wirelink does it, but we can detect a wrong response to a
DebugLinkGetState and skip it.

instead, download it to a temp location, as the Prophecy (and authors of
urlretrieve) foretold