fix: default ftINIT MILP solving to single-threaded Gurobi

[edkerk]

Summary

Fixes #607. ftINIT can fail with "Model is infeasible" / "Failed to find good enough solution within the time frame. MIPGap: Inf" for some cell types (e.g. vascularSmoothMuscleCells, pericyte) when building context-specific models from Human-GEM v2.0.0, while the same pipeline succeeds with v1.19.0.

As diagnosed by @mfcesur in #607, the cause is multi-threaded Gurobi solving the ftINIT MILP non-deterministically and spuriously returning infeasible. Forcing single-thread (Threads = 1) makes all cell types reconstruct, and is the more robust setting across Gurobi versions.

RAVEN already forces single-thread Gurobi, but only inside parallel workers (if ~isempty(getCurrentTask) in optimizeProb.m). In a normal serial ftINIT run getCurrentTask is empty, so Gurobi defaults to all cores — the failing condition.

Changes

• INIT/ftINIT.m — default Threads = 1 for the per-step reconstruction MILP, set in the existing param-defaulting block (overridable via the step's MILPParams).
• INIT/ftINITFillGapsMILP.m — same default for the gap-filling MILP.
• INIT/INITStepDesc.m — document the new default/override on the MILPParams property.

The default is overridable: set Threads in a step's MILPParams (e.g. 0 = all cores) and it is respected (user params applied last via structUpdate). It is solver-safe: glpk is blocked for MILP, scip ignores params, and cobra only acts on Threads with a Gurobi backend. The existing getCurrentTask safeguard is retained — it also covers non-ftINIT parallel solves such as randomSampling.

Testing

• checkcode clean on all three files.
• tinitTests ran locally, but the ftINIT MILP test (testftINIT_T0001) was skipped on the dev machine (no Gurobi/SCIP installed); CI (with Gurobi) exercises it.
• @mfcesur — could you confirm vascularSmoothMuscleCells/pericyte now reconstruct with default settings (no manual Threads edit)?

[mfcesur]

@edkerk, I tested the updated code and confirm that the issue is resolved on my side. I reran the reconstruction with the default settings (no manual Threads modification) for both problematic cell types and the Gurobi version/threshold combinations that previously failed.

With the new default single-threaded MILP setting introduced in this PR, both cell-type models were reconstructed successfully.

Thank you very much for the fix.