Fix CSP solver for XLinks

[jgraley]

See #193: We need to be able to extract this info (xlinks to master couplings) from the solution...

• so GetNextSolution() needs to deduce Values (=XLinks) for all variables including FORCEDs
• and that means they aren't really "forces" if they can change!!
• For each one, there needs to be a constraint that doesn't really correspond to any of the existing ones. It only needs to do the coupling (equivalence) check and should already know which X node to check against.
• See #170 - If agents are to be able to customise their coupling behaviour, then we need the agent for the master-coupled agent, and a SystemicConstraint for it, and a special mode that says "only check the coupling, and always against this X"

FREE/FORCED as a variable flag will have to go, since the new mode is on constraints themselves. Instead we want FREE/LOCATED on SystemicConstraint. A LOCATED SystemicConstraint:

• won't report any child links/variables
• is told where it is at about the time of SetForces() presently
• won't ever run eg DecidedQuery(), but only run coupling checks.

Note: while distinct pattern links are being supplied as variables (see PopulateForSolver()) and distinct X links are apparently being supplied as Values (see StartCSPSolver()), nothing is being done in SystemicConstraint to constrain couplings, which are now free to be non-equivalent due to the distinct variables. This is bug.

So:

• SystemicConstraints should be aware of having multiple incoming pattern links.
• For FREE SystemicConstraints, these will consist of exactly one keyer and zero or more residuals
• For LOCATED SystemicConstraints, all will be residuals
• SystemicConstraint::Test() will NLQ() etc if it has keyer, and then apply coupling restrictions to residuals (if there are any)
• The above logic then makes sense: a SystemicConstraint needs to be
  1. LOCATED and will be keyed before the solve, or
  2. FREE, has a keyer, so that it can key itself. Self-keying triggers NormalLinkedQuery() etc because the keying must be determined valid. Residuals cannot be considered until we have a key, and no further NLQ() is needed for these.

[jgraley]

Corrections:

• Omitted to mention that the CompareBy flag (LOCATION/EQUIVALENCE) can go
• We need to keep Freedom on variables (FREE/FORCED) because of the root x link
  • but this will now only be set to forced for the root link
• SystemicConstraint should be constructed as FULL or COUPLING. FULL implies "keying and residuals; does NLQ()"
• A FULL constraint requires, at planning time, the pattern link for the keyer
  • One day this requirement may disappear but for now we're doing asymmetric coupling restriction and engine must choose a keyer that won't MMAX, see #171
• All constraints require their residual plinks as well.

[jgraley]

Seem to be repeating https://github.com/jgraley/inferno-cpp2v/issues/169#issuecomment-712118851