Fix cuml parameter setting issues in UMAP/DBSCAN

[rishic3]

Fixed an issue where cuml parameters in UMAP were not being set due to a silent failure in _set_param, where these parameters could not be found in the _param_mapping. Mapping these params to themselves fixes this issue, as it is done in KNN.

DBSCAN had the same issue where cuml parameters were never set, but got away with this by using the Spark API to retrieve parameters and avoiding the use of cuml_params dict entirely.

Changes:

• Updated both DBSCAN and UMAP to include identity mappings for cuml parameters in _param_mapping.
• Updated DBSCAN to use the cuml_params dictionary for initialization to be consistent with the other algorithms.
• Updated the _param_mapping description to avoid future confusion for cuml algorithms without a PySpark equivalent.
• Added tests for DBSCAN and UMAP to check that cuml parameters are properly set using both the Spark param API and the cuml_params attribute, and test under non-default cuml parameter settings.

This PR also resolves Issue #749, as random_state is now properly set, making dataframe sampling deterministic.

[lijinf2]

Thanks for the PR! Looks pretty good.
Just one minor comment.

[lijinf2]

build

[eordentlich]

@lijinf2 should @rishic3 add similar tests to this PR for other algo params?

[lijinf2]

Yeah, that's a good idea.
@rishic3 Can you add similar checking to other algorithms? Each algorithm should have a test_params and we can add "assert est.cuml_params == cuml_params" to the test_params.
Also, is it possible to fix this in case estimator code does not include the "cuml_param": "cuml_param" mappings?