AI assistant: absorb a long paste instead of refusing it

Paste a long spec into the assistant today and it tells you the message is too long. That is better than what it used to do, which was nothing at all, but it is still a refusal. The person pasting fifteen thousand characters of requirements is not making a mistake; they are asking the clearest question they have.

Claude and Slack both take the paste instead. It becomes an attachment on the message, the box goes back to empty, and you type your actual question above it. Nothing is refused and nothing is lost.

We would need somewhere to keep the pasted text, a way to send it to Apollo as context rather than as the message, and a chip in the input showing what is attached with a way to remove it. The run attachments already give us most of the shape.

This supersedes the character counter added for https://github.com/OpenFn/lightning/issues/4883, which was the honest thing to do while we could not absorb the paste.

Start by reviewing the existing run attachments and the character counter described in issue #4883. Trace how pasted content reaches Apollo and how the assistant input represents attachments. Done means a long paste is retained as an attachment, the input can show and remove it, and Apollo receives it as context while the user’s message remains separate.