[WIP] ABLASTR: MLMGOptions for computePhi

[ax3l]

Prepare ImpactX for the breaking ABLASTR change in BLAST-WarpX/warpx#7228.

That PR replaces the four loose MLMG solver arguments of
ablastr::fields::computePhi with a single ablastr::fields::MLMGOptions
struct, which additionally exposes the bottom solver and the coarse-level
(agglomeration/consolidation/coarsening) controls.

Do not merge as-is. This is a work-in-progress PR so we can validate the
migration early and land it quickly once upstream is in. Before merging, the
first commit has to be replaced by a regular dependency update that resets
ImpactX_ablastr_repo/ImpactX_ablastr_branch to BLAST-WarpX/warpx and a
development commit that contains #7228.

Changes

• cmake/dependencies/ABLASTR.cmake: temporarily track
  roelof-groenewald/WarpX @ mlmg_solver_options (marked with a TODO).
• src/particles/spacecharge/PoissonSolve.cpp: fill an
  ablastr::fields::MLMGOptions from the existing defaults and algo.mlmg_*
  inputs and pass it to computePhi.

computePhi is the only ImpactX call site touched by #7228 — ImpactX does not
use computeEffectivePotentialPhi or getMaxNormRho, and the remaining
ABLASTR delta is additive.

No user-facing change

The algo.mlmg_relative_tolerance, algo.mlmg_absolute_tolerance,
algo.mlmg_max_iters and algo.mlmg_verbosity inputs and their sim.mlmg_*
Python properties are unchanged. The MLMGOptions members that ImpactX does
not set (bottom solver, bottom verbosity/iterations/tolerances, max coarsening
level, agglomeration, consolidation, final smoothing sweeps) default to the
same AMReX defaults that ImpactX used implicitly before, so the solve is
unchanged. Existing space-charge tests provide the coverage.

Testing

Built against the #7228 branch (OMP, MPI, FFT, double precision) with the
AMReX commit ImpactX currently pins — no bump needed, no new warnings.
The full CTest suite passes (201/201), including
cfchannel_spacecharge_mlmg, cfchannel_spacecharge_fft,
expanding_beam_mlmg, expanding_beam_fft, kurth_10nC_periodic,
expanding-fft-2d and chicane_csr.

Follow-ups

• Optionally expose the new bottom solver and coarse-level controls as
  ImpactX inputs — they gave a 16% runtime win on 8 A100s upstream.
• #7228 also lands ablastr::warn_manager::WMClear(), which would let ImpactX
  reset the ABLASTR warning state between ImpactX re-creations.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U98bFCQjdzryCL3idcxWgh

[codspeed]

Merging this PR will not alter performance

✅ 70 untouched benchmarks

---

_{Comparing ax3l:topic-ablastr-mlmg-options (f81e473) with development (75a7d5e)}