* Add visual-metrics docker type.

* Add required browsertime toolchain fetches.

* Add browsertime tests for technical and visual metrics.

* Run browsertime tests in a cron task.

* Run visual metrics on all browsertime tests.

* Use spaces instead of tabs, and resolve visual-metric nits.

* Enable browsertime on pull request for testing.

* Restrict PR tests to amazon on browsertime.

* First attempt using multi_dep.

* Add a primary dependency to browsertime.

* Try by not popping.

* Debug prints.

* Make one grouping per browsertime task.

* Try without the multi_dep transform.

* Delete dependent-tasks in visual-metrics transformer.

* Update setuptools installed and copy run-on-tasks-for.

* Use get when getting run-on-tasks-for.

* Add new pinned requirements.

* Try it.

* Set run-on-tasks-for properly.

* Remove print statement.

* Remove single_dep loader, and print statements.

* Remove run-on-tasks-for testing setting.

* Restart testing, and set user to root in visual-metrics Docker.

* Remove testing settings.

* Remove fetch-content from Docker.

* Change attributes grouping method.

* Run all tests as a check.

* Undo testing changes, and fix a bad test name.

There were a couple of errors in the original pull request for this bug:
- The raptor tasks had duplicated dependencies.
- The filename of the minidump_stackwalk artifact was incorrect.

The mozilla-central ffmpeg toolchain (used in Bug 1613479/#8780) has an
artifact that does not start with `public/build/` (it starts with just
`public/`). To easily support that, we want to set the `toolchain-artfiact`
attribute for that task.

We switched `index_search` to be a job-type in
[taskgraph](https://hg.mozilla.org/ci/taskgraph/rev/932b4cf48cc785c31b03f0b9f2cba654bc67bc00)
which pass through `attributes` from the task definition (and is also more
future proof for adding fenix-specific toolchain tasks). This updates taskgraph
and the consumers of the current toolchain.