MAINT Simplify filter_bank.py

[lostanlen]

closes #825
fewer lines, shorter docs, faster tests

• remove P_max=5, eps=1e-7, and normalize="l1" from Scattering1D object
• this lightens the kwarg lists of scattering_filter_factory and calibrate_scattering_filters
• unify morlet_1d and gauss_1d algorithms
• delete get_normalizing factors

supersedes #831 and #832
many thanx to @MuawizChaudhary

[codecov-commenter]

Codecov Report

Merging #833 (370f4d9) into dev (45afade) will decrease coverage by 0.19%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##              dev     #833      +/-   ##
==========================================
- Coverage   88.05%   87.86%   -0.20%     
==========================================
  Files          64       64              
  Lines        2311     2282      -29     
==========================================
- Hits         2035     2005      -30     
- Misses        276      277       +1     

Flag	Coverage Δ
jenkins_main	87.86% <100.00%> (-0.20%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
kymatio/scattering1d/frontend/base_frontend.py	98.55% <ø> (-0.07%)	⬇️
kymatio/scattering1d/filter_bank.py	99.27% <100.00%> (-0.73%)	⬇️
kymatio/scattering1d/utils.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 45afade...370f4d9. Read the comment docs.

[lostanlen]

My not-so-secret plan after this:

• remove adaptive_choice_P, make it part of morlet_1d (with appropriate comments to explain the formula)
• move periodize_filter_fourier to scattering1d/utils.py and remove its default value n_periods=1
• remove compute_sigma_psi, make it part of compute_params_filterbank (with appropriate comments)
• move compute_xi_max to scattering1d/utils.py
• remove calibrate_scattering_filters, make it part of scattering_filter_factory (i.e., scattering_filter_factory calls compute_params_filterbank directly)

Once that is done i think we'll be able to say that issue #513 is solved ;)
All that will remain in filter_bank.py will be:

• morlet_1d
• gauss_1d (which itself calls morlet_1d with xi=None)
• get_max_dyadic_subsampling (called twice by compute_params_filterbank)
• compute_params_filterbank (called twice by scattering_filter_factory)
• scattering_filter_factory

This will open an avenue towards modularizing filterbank construction, by switching to other wavelets (Morse/Gammatones/Meyer/other) and other compute_params_filterbank functions (the one we have, introduced by @janden, has a linear portion and an exponential portion, but in other cases we might want to guarantee that the xi's follow a geometric progression and have the sigma's be an affine function of xi's, as in ERB/"VQT" auditory filterbanks).

[MuawizChaudhary]

Looks good to me.

[lostanlen]

Rebased

[MuawizChaudhary]

LGTM

EDIT: Actually, the tests are failing. PLS fix

[lostanlen]

Fixed. That was my mistake in rebase