On warm start, provide `fmatch()` w/ capacity-feasible `x1`

[benthestatistician]

The relax algorithm certainly expects to be given a flow (x1) that's capacity feasible and satisfies CS at each arc. A flow of all 0's, as we provide now, does not do this, except perhaps in rare cases.

Unclear whether the B & T Fortran code does anything to compensate for provided x1's that don't have the right relationship to provided node prices. In relax4 source code, x is described as an "output parameter", not an input parameter. But so are reduced costs, and the solver certainly responds to reduced costs as input parameters (per @adamrauh's ugrad thesis).

[benthestatistician]

After additional review of the relax4f source -- see in particular "NOTE 2" and following -- it's clear that:

• In the manner in which we're now calling it, it doesn't care what flow x1 we give it, as it's settling on its own CS-satisfying flow during initialization phases 1 and 2;
• There may be a speed advantage to be had by informing it of the flow with which a prior run terminated, but securing that advantage would adjustment to manner in which we can the Fortran solver.

Specifically, we'd have to find a way to set the REPEAT common parameter to .TRUE., and also attend to additional issues noted in source code comments under "INITIALIZATION PHASE I". All that being beyond the scope of current activity, I'm closing out this issue.