Rework `get_duplicate_forecasts()` to S3 to avoid hard-coding columns and types

[nikosbosse]

          Again if we can rely on the input being a forecast object we can make this s3 and avoid a new class requiring hard coded changes to the package

Originally posted by @seabbs in #837 (comment)

[seabbs-bot]

Implementation plan (updated)

Problem

get_duplicate_forecasts() hard-codes c("sample_id", "quantile_level", "predicted_label") to determine which columns identify a unique row for duplicate detection. Any new forecast type with a different type-specific column requires modifying this function.

Approach

1. Add exported S3 generic get_forecast_type_ids()

Returns the type-specific columns (beyond the forecast unit) that identify a unique row. The default returns character(0) so only types with actual ID columns need a method:

Class	Returns
forecast_quantile	"quantile_level"
forecast_sample	"sample_id"
forecast_multivariate_sample	"sample_id"
forecast_nominal	"predicted_label"
forecast_ordinal	"predicted_label"
default (binary, point, etc.)	character(0)

Custom forecast types just define a get_forecast_type_ids method.

2. Keep get_duplicate_forecasts() as a plain function

No S3 dispatch needed on this function itself. It calls get_forecast_type_ids(data) internally.

3. Add type argument for raw data.frames

Users diagnosing duplicates on raw data before creating a forecast object can pass type (e.g. "quantile", "sample"). Internally constructs a temporary forecast object via new_forecast() to dispatch correctly.

Calling on a plain data.frame without type is deprecated but preserves the old column-detection behaviour for backwards compatibility.

4. Improve check_duplicates() error message

Includes the detected forecast type in the suggested get_duplicate_forecasts() call, e.g.:

Use `get_duplicate_forecasts(data, type = "quantile")` to identify duplicate rows

PR

Implemented in #1148.

This was opened by a bot. Please ping @seabbs for any questions.