Instead of handling a specific owner of the left brace (previously, it
was a Term.Match only), let's instead look for infix when we encounter
the right brace instead, and ensure that the next token is not the op
of an infix.

This also handles the case when Term.Match was not an LHS but in the
middle of an infix chain.

In older versions of scalameta, `extends` was not part of template, and
we had to identify the tree just preceding it.

Also, make sure to handle cases when the template is empty (in that case
there's no reason to break before a non-existent `extends`).

Historically, the parser returns `Type.Name("Unit")` for declarations
or definitions using the procedure syntax (i.e., no explicit `: type`).

However, that tree occupies no tokens and hence needs to be ignored.

Also upgrade scalatags.