Implement dielectronic recombination suppression calculation

[wtbarnes]

Fixes #233

TODOs

• Reconcile differences with IDL and published result
• Move to Ion?
• Implement special cases for $p_{N,0}$ in the lower-temperature correction.
• Tests
• Comparison tests with IDL
• Vectorize over arrays of density?
• Finish docstring
• Decide on how this gets integrated into actual DR calculation (could be left to another PR)
• More guidance from Ambiguity in implementation of dielectronic recombination suppression factors chianti-atomic/chianti-idl#17

[wtbarnes]

Here's the current comparison with the IDL code:

There are some minor differences, but given the differences in implementation, this isn't too surprising. In particular, it doesn't seem like the IDL implementation transitions smoothly to 1 for the H- and He-like ions.

What is somewhat surprising is the differences between what is implemented here and the published result (Fig. 2 of Nikolic et al 2018):

In particular, the O 4+ (Be-like sequence) appears very different to what is published. I'm not sure why this is or if there is a special case for $N=4$ that I'm not accounting for.

[wtbarnes]

The updated comparison (solid is fiasco, dashed is IDL) is below:

The H and He-like are now much closer to what is in the Nikolic paper. The Be-like suppression (for fiasco and IDL) is still quite different.

[wtbarnes]

I've now just made this a private method and I'm going to add the docstring info for this feature to the dielectronic recombination rate in a future PR as the user interface for this functionality is going to be via the actual rate calculation.

[pryoung]

I made some measurements from Figure 5 of Nikolic et al. (2018) to compare suppression factors at x=12.5 for the oxygen ions. I get

            IDL     Nik
  o_2     0.523   0.524
  o_3     0.237   0.237
  o_4     0.209   0.212
  o_5     0.551   0.618  **
  o_6     0.414   0.418
  o_7     0.595   0.594
  o_8     0.722   0.718

The agreement is good except for o_5. I think the Nikolic plot is wrong. If you check the math for the activation density (x), then the suppression factors for both o_5 and o_6 are expected to deviate from one at about the same density, and this is true for the IDL code, but not for Nikolic.
I've attached my version of the Nikolic plot with the latest versions of the IDL routines.

[pryoung]

For the apparent problem for He-like oxygen, I computed the suppression factors on a finer density grid, and I get the plot below. Both the o_7 and o_8 curves are smoothly varying around x=10, so this seems to be OK.

I compare the suppression factor values at x=10 for IDL and Nikolic below.

       IDL  Nikolic
o_7   0.958  0.937
o_8   0.986  0.973

There are small but significant differences. I think we can live with these, though.

[wtbarnes]

Thanks @pryoung. I did a comparison with your most recent changes and get very good agreement for all of the relevant isoelectronic sequences now,

There is a small offset in the Be-like sequence but its below 1% relative difference so I'm not that concerned,