Improve the EH transfer function by updating the zd and sh_d fits

[MinhMPA]

The sound horizon in EH 1998 fit is only accurate up to 2%. This is easily noticeable when one considers BAO quantities like DA/sh_d. This (in)accuracy further affects the transfer function.

The issue ultimately boils down to the fit for redshift at drag epoch, z_d given in EH 1998. So one can either improve on that, or directly on the sound horizon at drag epoch, zh_d. There are plenty literature on this, here I picked the z_d and sh_d fits from arXiv:2106.00428, which should provide sh_d at sub-percent accuracy (within LCDM).

[EiffL]

@claude can you review this PR and add a test of the new functions?

[claude]

Claude encountered an error —— View job

Failed with exit code 128

---

I'll analyze this and get back to you.

[MinhMPA]

@EiffL Not sure what the problem with the Claude review is. But since I started this PR, I have moved away from the fit in this PR to another one more well-behaved and much closer to class r_s(z_d)` output: Eq.~(16) of https://arxiv.org/pdf/1411.1074. Can update this PR if you could tell me where it currently fails.