add banded solver, TDC Y-face seed, skip cox if include_mlt_corr is F

[Debraheem]

This pr adds:

1. a matrix solver option with hydro_matrix_solver = 'banded' support and solver plumbing. This is the same banded solver RSP uses, just repurposed for star. It is significantly faster than bcyclic for pulsation models, or really anything that is on a smaller mesh. Bcyclic can parallelize more and remains a better all purpose generalized solver for MESA.

2. use_TDC_Y_face_seeded_newton plus the Y-face start/previous-iteration seed path.
   Small supporting TDC plumbing. use_TDC_Y_face_seeded_newton let's TDC skip the bisection logic and newton iterate to get a solution faster if the solution hasn't moved very far. This is helpful for zones where convection is not changing very much and another full root bisection can be wasteful. This control in my testing reduced the number of iterations in most TDC zones in half. This should be tested further, but this option should reduce the time spent in TDC ~ 1/2 . I have seen 5-10% speed up in some models during testing, but the speed up only becomes important when the timestep is limited by tdc/mlt and not eos/kap or other evolve step related overhead.

3. along with 2, when include_mlt_corr_to_TDC = .false., we now skip the call to Cox inside TDC, since we don't need mlt Gamma. This is one less extra mlt call inside tdc per zone per iteration.

[VincentVanlaer]

Is there a test case you could enable this in? That way we don't break this option.

[Debraheem]

Is there a test case you could enable this in? That way we don't break this option.

Yes, I will be using this in the pulsation models I introduce to the test suite for TDC.

[Debraheem]

RSP2 will likely use it as well, once it's up and running (the banded solver). The TDC_Y_face_seeded_newton can be turned on in any test case we want.