Data.Graph.bcc is not efficient

* The `collect` function recursively `concat`s on a tree, making the time complexity quadratic. This can be avoided using difference lists.
* The `do_label` step builds some unnecessary intermediate trees, similar to the `dfs` we had (#882).

I'm not aware of common use cases for `bcc`, so I'm not sure if this affects anyone.
But as long we have it, we should make it efficient.

I can send a PR.

Start with Data.Graph.bcc, focusing on the collect and do_label steps, and compare them with the dfs work referenced in #882. Done means bcc avoids the reported quadratic concatenation and unnecessary intermediate trees while preserving its existing behavior.

Written by the indexing model from the issue text.