Tabular Deep Learning parameters

Parameters that need to be added

Argument	Model	Type	Why new	Existing alternative
attention_type	saint	qualitative ("row", "col", "both")	No qualitative param for attention mode exists	none — genuinely new
dropout_last	saint	quantitative, proportion [0,1)	No param for final-layer dropout; matches the dropout_attn()/dropout_embedding() precedent of per-location dropout params	could reuse dropout(), but precedent favors a new param
dropout_hidden	saint	quantitative, proportion [0,1)	Saint has no generic `dropout` arg; this is its hidden-layer dropout	could reuse dropout() (most defensible reuse), or new per precedent

For attention type it is "column", "row", or "both

Yes sir, that's what's in the PR:

Line 25 in a16c809

values_attention_type <- c("column", "row", "both")