fix(clone): theory overrides of nested or overlapping theories

[strub]

Two problems with theory overrides theory X <- T in clones.

An override of a nested theory, theory B.A <- T, was recorded under
its last component, at the top level: an unrelated theory A of the
cloned theory lost its import, and a second override theory A <- T'
failed with an assertion failure. Theory overrides are now recorded
under their full path, and a duplicate one is reported as such.

A theory override together with a more specific override, as in:

clone Leaf as L with theory B.A <- Arr, theory B <- B0.

failed with an assertion failure, or reported a duplicate override
when given in the other order. Theory overrides are now processed
after the other overrides, the most specific first, and leave alone
the items that already have an override: the more specific override
wins, whatever the order.

In particular, a theory B produced by a clone that overrode its
sub-theory A with theory A <- U can now be threaded through
further clones with <-. That sub-theory was cleared, so that
theory B <- B alone fails, as B.A has no target; the clones give:

theory B <- B, theory B.A <- U

where <= was needed so far, at the cost of a copy of the threaded
theory in every clone.

Also, a clone now replays a theory alias with its substituted target,
instead of the original one.