fix type in featureselection converters

[Felipedino]

Problem

The feature selection converters (SelectKBest, SelectPercentile,
SelectFdr, SelectFpr, SelectFwe, GenericUnivariateSelect and
VarianceThreshold) hardcoded their get_output_type to always return
Float (float64). Since these converters only drop columns and never modify
the values of the retained ones, this corrupted the type: an integer column was
reported as float64 even though the underlying data stayed integer.

Solution

Preserve each retained column's original type:

• Added to the base class FeatureSelectionConverter a fit that remembers the
  input types and a get_output_type that returns the original type per column
  (falling back to float64 only when the type is unknown). This covers the 6
  scikit-learn selectors.
• Removed the duplicated get_output_type (and the now-unused DashAIDataType
  import) from the 6 selector files, which now inherit the behavior.
• Applied the same fix to VarianceThreshold (same bug, same "only drops
  columns" nature).

Key detail

Types are captured in fit and not in transform: scikit-learn
(_SetOutputMixin.__init_subclass__) automatically wraps any transform
defined on a subclass of a sklearn transformer and would coerce the output back
into a pandas DataFrame. fit is never wrapped, so it is the safe place and it
always runs before transform.

Verification

• End-to-end test: integer columns stay int64, float columns stay float64,
  and the declared type matches the underlying arrow data.
• ruff check clean.
• Existing converter tests pass (including test_base_converter_metadata.py).

Modified files

• DashAI/back/converters/category/feature_selection.py
• DashAI/back/converters/scikit_learn/select_k_best.py
• DashAI/back/converters/scikit_learn/select_percentile.py
• DashAI/back/converters/scikit_learn/select_fdr.py
• DashAI/back/converters/scikit_learn/select_fpr.py
• DashAI/back/converters/scikit_learn/select_fwe.py
• DashAI/back/converters/scikit_learn/generic_univariate_select.py
• DashAI/back/converters/scikit_learn/variance_threshold.py