Warn the user when PDF text extraction returns nothing

When client-side PDF extraction produces little or no text, the message is

sent without the attachment and nothing indicates this happened. The model

then answers without the document, and the user has no way to tell the

difference between "the model read it and misunderstood" and "the model never

received it."

I only discovered this by instrumenting my own endpoint and confirming the

request body contained just my question text, with no attachment content of

any kind.

Request: a visible warning when extraction yields little or nothing, ideally

before the message is sent.

This is likely the cheaper of my two requests to implement, and it prevents

silent wrong answers regardless of whether raw PDF forwarding ever ships.