perf: `_find_all_targets()` has quadratic complexity because of `Git.is_tracked()`

[Suor]

Git.is_tracked() recreates a list of tracked files for each call and is called for each file walk_files() will find. Both lists are proportional to repo size, which makes _find_all_targets() ~ O(repo_size ** 2).

[efiop]

https://github.com/iterative/dvc/blob/0.77.3/dvc/scm/git/__init__.py#L236

[Suor]

There should be a way to access git index by name and thus implement Git.is_tracked() in O(1), making things linear overall.