ndcg refactor

[FelipeAdachi]

Description

This PR:

• Adds ndcg value expecations to the tests, not just counts
• Changes convert_non_numeric=True to be used for string columns (integers, floats in scores and bools should use the default convert_non_numeric=False )
• adds score_column support: when a score is available, like in this example, one should use score_column and target_column , and not prediction_column (see 2 last tests on the sklearn examples)
• Changes the logic to generate target column values to make it compatible with all scenarios
• Fixes prediction and ideal relevance calculation for non numeric case
• Handles DivisionbyZero edge case when idcg=0 (ndcg set to 1 if no relevant documents exist)
• If K is not passed, metrics will be calculated according to predictions cols's length (and metrics named accordingly - k is no longer omitted in the names when k is None)

For the Numeric case:

• If predictions+target or scores+target columns are both provided, they need to be of same length.

• If prediction_column is provided with target column, prediction col contains the rank of suggested items, starting with 1

• If only prediction column is provided, it is assumed that the order encodes the ranks of recommendations: first item in the list is the first recommendation. The value in the list encodes the relevance score

• I have reviewed the Guidelines for Contributing and the Code of Conduct.

[jamie256]

Left a few questions and minor comments. Thanks for working on these!

[jamie256]

Looks better, thanks!