docs: improve error message when trying to infer the dtype of Deferreds

[NickCrews]

See #11608 (comment)

[JonAnCla]

sorry for my ignorance: is there somewhere I can refer to to get an idea of circumstances that cause inference to be triggered? seems like its not when building a Deferred which makes sense, so presumably its when incorporating a Deferred into an expression bound to a concrete table or column?

[NickCrews]

honestly, I don't see/understand all the cases either.

This is one example that exercises it:

import ibis

t = ibis.memtable({"a": [1, 2, 3], "b": [4, None, 6]})
t.fill_null({"a": ibis._.b})

I found this by looking up call sites of dt.infer().

I'm not sure this is required/correct to use infer() here. I think probably we should actually be binding the values to the table first, and then we would be able to resolve their values.

[NickCrews]

So maybe we should actually frame this error message as "this is unexpected behavior, please file a bug with the ibis maintainers"?

[NickCrews]

I am trying to fix this particular bad implementation at #11698.

There could still be other places in the ibis codebase where you could trigger this.