Replacement fix

[jrmaddison]

Prior to this PR Replacement objects shared identity with originating variables in a number of ways:

1. Shared their tlm_adjoint variable ID (obtained via var_id) with originating variables. Intended and required.
2. Shared their UFL 'count' with originating variables. Intended, simplifies the code.
3. Compared as equal with originating variables in some cases. Not intended, seems fragile.
4. Shared their UFL count with other Replacement objects in some cases. Not intended, seems likely to lead to bugs.

The last of these is a consequence of Firedrake Constant and Function objects having their counts set independently, and would be resolved more easily by not replacing Constants at all -- since in most cases only scalar Constants work with the Firedrake backend.

This PR fixes this by ensuring that Replacement objects have their own UFL counts, meaning that 2., 3., and 4. no longer hold.

However

The order of var_replacement calls is dependent upon the behaviour of the Python garbage collector. While reference dropping occurs at specific points (controlled by the EquationManager), precisely which references are dropped and in what order depends upon details of garbage collection. This means that if we wait until the var_replacement call to instantiate a Replacement then their UFL counts might be ordered differently on different processes. which also seems fragile. This PR resolves this with an awkward partial initialization of Replacement objects, with the count set on originating variable instantiation, and the VariableInterface bound later on the var_replacement call. This PR resolves this by generating counts on originating variable instantiation.

[jrmaddison]

Some side-notes:

• Cofunction instances compare as not equal with themselves (via __ne__). This is worked around in this PR. Now fixed upstream.
• Given Cofunction and Cofunction subclass instances with the same count __eq__ is not even reflexive (one of the few behaviors not permitted in Python, https://peps.python.org/pep-0207) -- another case against letting replacements and originating variables share counts.

[jrmaddison]

Pending an update to the Firedrake docker container