[linfa-svm] Fix SVR nu parameter passing and rework SVR parameterization API

[relf]

This PR fixes and clarifies the SVR parameterization with regards to the original libsvm code
before:

• nu_eps(nu, eps) was wrongly wired: eps was used for nu and nu for c
• c_eps(c, eps) is confusing: eps refers is epsilon of the inner loss function while there is also an epsilon used for the solver
  after:
• nu_svr(nu, Option<c>) replaces deprecated nu_eps(), optional parameter default is 1. as in libsvm
• c_svr(c, Option<loss_eps>) replaces deprecated c_eps() second optional parameter default is 0.1 as in libsvm
• the epsilon of the solver is set through SolverParams::epsilon() as before

• Bump linfa-svm version to 0.7.2

Fixes #317

[codecov]

Codecov Report

Attention: Patch coverage is 20.00000% with 12 lines in your changes missing coverage. Please review.

Project coverage is 34.93%. Comparing base (936680a) to head (0192092).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
algorithms/linfa-svm/src/hyperparams.rs	7.69%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #370      +/-   ##
==========================================
+ Coverage   34.78%   34.93%   +0.15%     
==========================================
  Files          96       96              
  Lines        6324     6334      +10     
==========================================
+ Hits         2200     2213      +13     
+ Misses       4124     4121       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[relf]

Not sure to understand why codecov/patch is not happy. The missing lines should be covered by tests in regression.rs, so... @quietlychris, @xd009642, any idea?