fix backward feature selection for fixed_steps

[johnrushwrs]

Backward feature selection has some issues, namely:

• n_terms is incorrect as it should be the num_available_features - n_steps
• initial_state is incorrect as it should be the list of initial predictors in backward selection

@jonathan-taylor do you mind taking a review here? only tried to fix the fixed_steps case, but I imagine first_peak has the same issue as well.

Love the ISLP library and book by the way! I just found this issue while going through the exercises in chapter 6, 8.d was not possible to do with the current package. LMK if the fix looks good :)

[johnrushwrs]

Test output:

$ pytest -s  tests/models/test_selection.py::test_fixed_steps_backward
======================================================================= test session starts ========================================================================
platform darwin -- Python 3.12.7, pytest-7.4.4, pluggy-1.0.0
plugins: anyio-4.2.0
collected 1 item                                                                                                                                                   

tests/models/test_selection.py selected ['A', 'D', 'G']
selected ['A', 'B', 'C', 'E', 'G']
.

======================================================================== 1 passed in 1.66s =========================================================================