New feature: alternative definition of transfer function

[adrien-laposta]

This PR implements a new feature in soopercool which allows the user to flip between two alternative definitions of the transfer function.

It includes new fields in the parameter file in the transfer_settings block, tf_ordering and power_law_c_ell.
Transfer function estimation now relies on input Cl instead of pseudo-Cl of unfiltered simulations.

[adrien-laposta]

A few more words on how to write a parameter file. The transfer function section should be

transfer_settings:
  transfer_directory: /where/to/save/the/tf
  tf_est_num_sims: N
  sim_id_start: 0
  power_law_c_ell: /path/to/powerlaw/cell [format should be a npz file with key ["l", "cl"]
  tf_ordering: "MT"/"TM"
  filtered_map_dir:
    ftag1: /path/to/maps
    ftag2: /path/to/maps
  filtered_map_template:
    ftag1: {pure_type}_{...}_{id_sim:04d}_{...}.fits
    ftag2: {pure_type}_{...}_{id_sim:04d}_{...}.fits

[adrien-laposta]

Now added a tutorials directory in the repo to start accumulating small, self-contained notebook numerical breakdown of what we do within the software. The first one is about transfer function definitions and the related linear algebra.

[adrien-laposta]

Additional comment:
I did not delete the previous approach relying on the computation of both unfiltered and filtered power law spectra for TF. compute_pseudo_cells_tf_estimation.py will still compute both even if they're not used in the current script to get the transfer function.

I'd like to get your opinion, should we completely remove it ? I have the feeling that letting too many options will lead to confusion. If we need it, we'll still be able to come back to a previous version and compare.

[kwolz]

Sound good to me, happy for this to merged!

[damonge]

Just a few small comments. Looks good to me.