Only update the _variables_to_constriants_mapping dict when new_name !=
old_name

Only _round_primals_to_bounds if solution is optimal

Default is set to 1e-9

E.g. numpy and sympy

* feat: add get_linear_coefficients method

Abstract non-implemented method

* feat: implement glpk get_linear_coefficients method

* fix: typo

* feat: implement get_linear_coefficients in cplex

* feat: implement get_linear_coefficients in gurobi

* feat: implement get_linear_coefficients in scipy

* test: add tests for get_linear_coefficients

* test: more tests for get_linear_coefficients

* fix: flake8

* fix: missing arguments in test

* fix: make scipy consistent with other interfaces

the get_linear_coefficients and set_linear_coefficients methods now
raise an exception if the objective/constraint doesn’t belong to a model

Some cplex versions don’t have all the statuses that optlang tries to
import, which would result in a crash.

For range constraints, rhs is now equal to lb

This is due to a bug where gurobi can get stuck

* checking if __del__ is needed

* testing against miplib

* using problems from miplib2003

* needs to implement variable comparison for problems that converge

* more problems
runtime: 10 min per problem

* fixed milp test value (same as in devel)

* delete problem from memory only if it exists (__del__ method)

* add a .keep file to keep the miplib2003 directory on tests/data

* miplib test with cplex - some problems also fail...

* need to implement a parser for .sol files

* set 9 min limit for optimization to run, otherwise travis halts after 10 min without output

* refactor: tests and move into slow_tests top directory

* style: flake8

* fix: no point in searching ImportError for 'cplex'

* test: add slow MILP tests for gurobi