implementing automatic smoothening parameter selection

[amanpandey2587]

I have implemented the following changes after referring the document provided in the issue :

• I have added higher-order variance and link derviatives to distributions and link functions in order to support smoothness derivative calculations.
• I have made a new file named pygam/smoother_optimizer.py with QR/rank utilites, finite difference gradient/Hessian helpers and a safeguarded modified-Newton step function.
• I have also implemented fit_auto_smooth in GAM for Wood-2008 style outer-loop smoothing selection around the existing PIRLS inner loop which included dense penalty helper and objective wrapper.
• I have kept backward compatibility while storing optimization metadata (lam_opt_result, tuned λ) and wiring lambda handling utilities.
  -Finally I added comprehensive tests for derivative correctness, QR sanity, finite-difference helpers, and auto-smoothing on Gaussian, Poisson, and logistic concurvity scenarios.

** Solves issue ** #192

Tests : There are two tests included in

• pygam/tests/test_outer_lambda.py (outer-loop λ search coverage)
• pygam/tests/test_smooth_optimizer.py (derivative helpers and auto-smoothing integration)
  Below is the result showing their successful passing :