Highs Solver support

[schel337]

Adds support for highs LP solver. Seems to be a bit faster than the cuOpt simplex.

[leiwenpua1251-create]

Hey @schel337, think there might be an issue in the high_flux handling in solve(). It reuses _base_ub[col] as the upper bound when pinning a column's lower bound near its optimum:

self.problem.changeColBounds(col, limit, self._base_ub[col])

For synthetic exchange columns (added_secretion/added_uptake), _base_ub[col] is 0 (they're pre-allocated disabled). So when limit > 0, this sets bounds to [positive, 0] — infeasible by construction, regardless of the actual LP. Reproduced it on RECON1 (metabolite dopasf[c]): a feasible secretion_max of ~1.0 followed by an infeasible high_flux solve at limit=0.95.

My thinking: track a separate upper bound for high_flux (real max for structural reactions, model.maximum_flux/EXCHANGE_LIMIT for synthetic exchange columns) instead of reusing _base_ub.