Support intra-horizon topology changes

[m-bossart]

Today a component is either in the model for the entire horizon or absent
entirely. There is no way to represent a line going out of service midway
through the day.

Current state

• Availability is a build-time boolean (get_available_components → IS.get_available).
• NetworkModel.network_matrix is a single matrix for the whole horizon.
• DCP ohm law uses a constant b for every t; PTDF uses one column per branch for every t.
• PSY outage attributes (PlannedOutage, FixedForcedOutage) exist but POM never reads them.

Note that driving a BranchRatingTimeSeriesParameter to zero does not emulate an
outage — impedance is unchanged, so the line still carries flow in the PTDF/ohm law.

Proposed approach (exogenous, known outage schedule)

• Resolve PSY outage attributes into a per-timestep availability mask.
• DCP: skip the NetworkFlowConstraint row and the branch's ActivePowerBalance
  contribution on outaged steps.
• PTDF: reuse the MODF machinery from the security-constrained path. modf_matrix[arc, spec]
  already returns a full injection→flow row against the same injection variables, and
  VirtualMODF shares the VirtualPTDF factorization, so states are Woodbury updates
  rather than new PTDF builds.

Known gaps

• MODF expressions are currently built only for monitored arcs; base-case coverage needs
  every rated branch × state. Each (arc, state) is a serialized KLU solve — needs benchmarking.
• The outaged branch itself needs its flow zeroed and rating constraint dropped.
• Islanding: MODF handles it at the matrix level (is_islanding, bus_island_labels), but
  the power balance does not — subnetworks is one partition applied to all t. Affects DCP equally.
• Outage registration currently skips silently on failure and falls back to the unmodified
  PTDF row. Acceptable for SC monitoring; on this path it would silently solve the intact
  network and must become a hard error.
• Dual/LMP path not yet traced for single-matrix assumptions.

Out of scope

Endogenous topology (optimal transmission switching) — needs branch binaries and a
disjunctive ohm law, and is structurally incompatible with PTDF.