fix: remove_redundant_facets mutates caller's lb/ub arrays (issue #83)

[xovishnukosuri]

Problem

remove_redundant_facets in dingo/pyoptinterface_based_impl.py modifies the lb and ub arrays it receives in-place. It replaces values with sys.float_info.max / -sys.float_info.max when marking redundant bounds and near-zero-width constraints.

PolytopeSampler.get_polytope() passes self._metabolic_network.lb and self._metabolic_network.ub directly into this function. After the first sampling run those arrays on the model are corrupted. Any subsequent call — a second generate_steady_states, an FBA, or an FVA on the same model — sees bounds like -1.79e+308 instead of the original values.

This is the root cause of the UnboundLocalError reported in issue #83: the second FBA on the corrupted model returns a non-OPTIMAL status, optimum_sol and optimum_value stay at their zero-initialised defaults, and downstream code that expects a real solution breaks.

Fix

Copy lb and ub at the top of remove_redundant_facets so all internal modifications are confined to local arrays. Two lines added, no API change, no performance concern for normal model sizes.

Test

Added test_remove_redundant_facets_no_mutation to tests/fba.py. It snapshots the model's lb and ub before the call and asserts they are byte-for-byte identical after it returns.

Reproduction (before fix)

import dingo

model = dingo.MetabolicNetwork.from_json("ext_data/e_coli_core.json")
model.set_opt_percentage(90)
sampler = dingo.PolytopeSampler(model)
sampler.generate_steady_states()

model.set_opt_percentage(20)
sampler = dingo.PolytopeSampler(model)
sampler.generate_steady_states()  # UnboundLocalError before fix

Closes #83