Follow-up from "Draft: tor-keymgr: Initial implementation"

The following discussion from !1223 (merged) should be addressed:

• @Diziet started a discussion: (+1 comment)

  Maybe the dummy KeyMgr could have the same constructor API as the real one. But let's not add more stuff to this MR. If you agree, how about maybe a TODO or a ticket and we can do it later?

Also: the dummy KeyMgr should have its own error type.