fix(core): prevent mutation via deepcopy in get_params

[Aditya-9215]

Summary

Fixes issue where get_params() returns references to mutable nested objects,
allowing unintended mutation of internal model state.

Problem

Currently:

params = gam.get_params()
params["terms"][0].lam = 1000

mutates the original model.

This violates sklearn’s estimator API expectations and can break clone(),
GridSearchCV, and reproducibility.

Solution

• Updated Core.get_params() to return deep copies of parameters
• Prevents mutation of nested objects (e.g., Term, TermList)
• Fix is applied centrally at Core level so all subclasses inherit correct behavior

Tests

Added test:
test_get_params_does_not_mutate_model

Notes

Deepcopy introduces minor overhead but ensures correctness and API compliance.

🔗 Related Issue

Closes #522