hook up to MOI (WIP)

[chriscoey]

fixes #9
requires transforming symmetric matrix inputs/outputs using scaled svec form (would be nice if MOI still supported scaled triangle)

• contlinear tests
• contconic tests

[chriscoey]

this will be easier after #62 because we will have support for constraint cones

[chriscoey]

🎉 MOI conic tests are mostly passing except for

• SDP test failures (must be some bug in Hypatia's PSD cone functions)
• a couple of errors from Hypatia probably due to full row rank requirement on A unsatisfied (redundant constraints): need to fix presolve to ensure rank conditions are satisfied #16

I plan to fix the SDP test failures and then merge this, and work on #16 soon so that all the MOI tests pass

[chriscoey]

Actually SDP may be failing simply because MOI's PSD triangle is unscaled, whereas hypatia's is scaled (svec form)

[chriscoey]

excluding SDP tests until jump-dev/MathOptInterface.jl#531 is fixed

[chriscoey]

besides SDP/logdet/rootdet tests, which will be fixed by MOI bridges, only 3 test failures remain across contlinear and contconic tests, and they are only on LPs. 2 failures are due to redundancy or inconsistency of dual equalities, and 1 failure (linear15test) is due to having more equality constraints than variables (p > n).

both will be solved by #16.

[chriscoey]

restoring tolerances to 1e-4 cause a failure on rotatedsoc3, that might be solved by #65