Confirm post-contingency flows are correct when the outaged branch is itself a phase shifter

#225 added DC phase-shift equivalent injections (+b·φ at from, −b·φ at
to) to the nodal balance, with a matching −b·φ offset on each
monitored branch's own flow at
src/ac_transmission_models/security_constrained_branch.jl:580. Two
cases were never verified:

1. Outaged phase shifter. The injection vector handed to the MODF
   builder still carries the ±b·φ pair of every shifted arc, including
   the outaged one. If a contingency removes a phase shifter, its
   injections arguably should come out of the post-contingency vector. If
   they don't, post-contingency flows on all monitored branches are
   wrong — silently, with plausible-looking numbers.
2. DCP networks. add_post_contingency_flow_expressions! for
   DCPNetworkModel builds the vector via _dcp_nodal_injection_expressions
   (:705, defined :771) rather than using the ActivePowerBalance
   expression directly, so it may not contain the shift injections at all
   — while :580 still subtracts the offset.

May well be correct; it has simply never been tested. Note this is a
net improvement regardless — before #225, phase shifts were ignored
entirely in PTDF/MODF flows.

Acceptance: a test registering a contingency on a phase-shifting
transformer, comparing the monitored post-contingency flow against an
independent computation, under both PTDFNetworkModel and
DCPNetworkModel.