Hybrid-PIC: checkpoint the electron temperature under the electron energy equation

[clarkse-he]

The bug

With hybrid_pic_model.solve_electron_energy_equation = 1, the electron temperature
hybrid_electron_temperature_fp is the evolved state variable of the QDSMC
electron-energy equation — it carries the accumulated Joule heating, the Q_ei
equilibration and the advected entropy. It is not written to the checkpoint, so on
restart it is reconstructed from scratch twice over:

1. HybridPICModel::InitData fills it with the uniform <hybrid>.elec_temp, and
2. WarpX::HybridPICInitializeRhoJandB then overwrites it via
   CalculateElectronPressure(floor_density=true), the floored adiabat closure on the
   restored rho.

Every restart therefore silently discards the evolved electron thermal structure, with
no warning. A checkpointed run is not equivalent to an uninterrupted one, and the
divergence feeds straight back into the fields through Pe.

Source/FieldSolver/WarpXPushFieldsHybridPIC.cpp acknowledges this in a comment today
("T_e is not checkpointed either … evolved T_e structure is not preserved across a
restart"), but nothing warns the user at runtime.

Measured impact

On a 2D 48² Ohm's-law deck with the energy equation and Joule heating (checkpoint at
step 5, compared at step 10 — five steps of divergence), restart vs. uninterrupted,
relative max error:

field	error
Te	9.1e-02
Ez	1.7e-01
Ex	5.6e-03
jx	6.4e-03
Bx	1.4e-04
rho	2.0e-05

With this PR the same pair is bit-identical (max 1.0e-16, roundoff).

The fix

• Flag hybrid_electron_temperature_fp with checkpoint_restart when the energy
  equation is on. With the equation off it stays unflagged — there it is a pure
  diagnostic mirror of the closure, refilled every step, so checkpoints for the
  algebraic-closure path are unchanged.
• ablastr::fields::MultiFabRegister::read_restarts now skips a flagged field whose
  file is absent from the checkpoint instead of aborting, and returns the names it
  actually read. This keeps older checkpoints restartable. It is also a latent fix
  independent of this PR: the pre-existing checkpoint_restart field in
  NuclearFusionFunc.H makes a checkpoint written without fusion unrestartable with
  fusion enabled today.
• WarpX::InitFromCheckpoint sets m_te_restored_from_checkpoint when T_e came back on
  every level, and records a WarnPriority::high warning when the energy equation is on
  but the checkpoint has no T_e — in which case the adiabat re-seed still happens,
  exactly as before, so nothing regresses for existing checkpoints.
• Both T_e seeds are skipped when that flag is set, and HybridPICInitializeRhoJandB
  emits Pe from the restored T_e via QDSMCFillElectronPressureFromTe (with the
  ApplyElectronPressureBoundary + FillBoundary treatment grad(Pe) needs) instead of
  re-seeding.

This is the electron-energy-equation companion to #7049, which fixed the rho/J/Pe half of
the same restart path.

Testing

Kept source-only to stay light; verified locally against a restart-equivalence deck (see
comment below). The pre-existing non-energy-equation restart pair is untouched and still
matches the stored benchmark test_2d_ohm_solver_checkpoint_picmi.json to 0.00e+00.

Happy to add a test_2d_ohm_solver_ee_checkpoint_picmi{,_restart} pair to
Examples/Tests/ohm_solver_restart if reviewers would like the coverage in CI.

[clarkse-he]

Verification, for the record — run locally rather than added to CI to keep this PR light.

I built a restart-equivalence deck alongside the existing ohm_solver_restart pair: same
grid, drive and restart-hostile ingredients (parsed analytic initial B, split-field
external vector potential with a time envelope, density floor), plus
solve_electron_energy_equation and include_joule_heating with
plasma_resistivity = 1e-3, so T_e reaches ~2.3× its seed value by the checkpoint step.
2D, 48², 10 steps, checkpoint at step 5, compared at step 10 — 2 MPI ranks,
OMP_NUM_THREADS=1, CPU/OMP build.

Before the fix, the restarted arm diverges from the uninterrupted one:

field  Bx: rel err = 1.417e-04
field  By: rel err = 1.543e-05
field  Bz: rel err = 1.823e-05
field  Ex: rel err = 5.600e-03
field  Ey: rel err = 3.843e-03
field  Ez: rel err = 1.680e-01
field  Te: rel err = 9.106e-02
field  jx: rel err = 6.381e-03
field  jy: rel err = 3.495e-05
field  jz: rel err = 1.834e-03
field rho: rel err = 2.012e-05

After the fix, analysis_default_restart.py passes at its 1e-12 tolerance — every
field is at roundoff:

field  Bx: 1.615e-17    field  Ey: 3.862e-17    field  jx: 0.0
field  By: 0.0          field  Ez: 1.036e-16    field  jy: 0.0
field  Bz: 0.0          field  Te: 1.043e-17    field  jz: 0.0
field  Ex: 3.005e-18                            field rho: 0.0

Also checked:

• Backward compatibility. A build with this fix, restarted from a checkpoint written
  before it (no hybrid_electron_temperature_fp file), runs to completion and emits the
  expected high-priority warning — read_restarts skips the absent field rather than
  aborting.
• No regression on the algebraic-closure path. The existing
  test_2d_ohm_solver_checkpoint_picmi pair still reproduces the stored benchmark at
  0.00e+00 max relative error, and its restart comparison is unchanged at roundoff.

The deck is a small variant of the existing inputs_test_2d_ohm_solver_checkpoint_picmi.py
and I can push it plus a benchmark JSON as a second commit if you'd rather have this
locked down in CI.