testport can publish a repository when it shouldn't.

[bdrewery]

Consider a bulk -c that fails using a .building directory, then a testport on 1 port is ran and succeeds. That testport will then publish the repository with only that 1 package and its dependencies intact.

I think testport may still be deleting the actual port it tested as well.

This comes back to the "partial build" problem that TRIM_ORPHANED_BUILD_DEPS was added in relation to.

Testport really needs to be more sandboxed on the package repository usage so that it doesn't publish partial builds or trim away expected packages.

[bdrewery]

Related to #617

[bdrewery]

This is trivial to fix with another package dir like .building but dedicated for testport and never committed after. From a user perspective I would prefer it did commit all but the port listed for testport but restore the old version of it. That should be trivial too.