Important bugfixes to QHACalc now that fixed-cell relaxations are carried out

[Andrew-S-Rosen]

This PR fixes additional bugs with the QHACalc, following up #158, in which I fixed a bug where the phonon calculations were being carried out on the scaled structures without any ionic relaxation carried out. It also closes #138 by adding support for checking imaginary modes.

Using the correct energies

The energies used for the EOS fit should be taken from the minimum energy structures at the fixed-volume relaxations. This was not done before but is now done.

Consistent settings

The fmax and optimizer should be used for all relaxations consistently. This was not done before but is now done by passing the arguments to all RelaxCalc calls.

Relax cell shape

While a fixed-cell volume relaxation should be carried out, the cell shape should be allowed to change. I have enabled this. In doing so, RelaxCalc takes a new keyword argument relax_cell_kwargs so that we can do FrechetCellFilter(constant_volume=True).

Imaginary modes

There is now a keyword argument imaginary_freq_tol to optionally check for imaginary frequencies.

Force tolerance

I also have better addressed #127. In short, the existing fmax default value is not appropriate for phonon calculations. The forces must be very small to carry out accurate phonon calculations. The standard in the literature varies anywhere from 1E-5 to 1E-8 eV/A. The default of 0.05 eV/A (while lower than 0.1 eV/A before) is not scientifically sound. People will be doing very bad science with this kind of default. For an example, in this MLIP paper, the forces are converged to 1e-8 eV/A. I have made fmax set to 1e-5 as a better default. I also increased max_steps to 5000 to ensure the calculations are reach convergence.

[Andrew-S-Rosen]

This is now ready for review. Tests pass locally, but the CI is messed up.

[rul048]

I'll take a closer look and try to fix the CI issues as soon as possible.

[rul048]

Thanks for the improvements and for addressing these issues with QHACalc. The changes look good and the fixes make sense scientifically. I have merged the PR. Much appreciated!