Requirements for scalar UDF preimage

`ScalarUDFImpl::preimage` allows defining "preimage" (range of input values) for function outputs, which can be used to rewrite binary comparison expressions.

The rewrite rules for equality (`=`) or inequality (`!=`) seem intuitive. But `f(x) <= c` would also be rewritten to `x < u` where `u` is the (exclusive) upper bound. (Here is the [code](https://github.com/apache/datafusion/pull/19722/changes#diff-3f015bb6ca85b8af8daaef1f0959e0b49ec6bcbcf675b9ea2d87d52a95893773).) I'm not sure if this holds in general. Does this assume that the function must be monotonically non-decreasing?

I was trying to see if the documentation mentions monotonicity for scalar UDFs but I couldn't find it. Let me know if I misunderstood anything!

Start with the ScalarUDFImpl::preimage entry point and the rewrite rules for equality and inequality referenced in pull request 19722. Check whether existing scalar UDF documentation addresses monotonicity or preimage requirements. Done means documenting the required assumptions and clarifying whether the <= rewrite is valid in general.

Written by the indexing model from the issue text.