docs: fix delete_old_messages example to properly demonstrate deletion effect

[Udit Sharma (uditstocks)]

Overview

The existing delete_old_messages example in the short-term memory documentation failed to properly demonstrate
the deletion effect. After deleting the earliest two messages, the agent could still recall the user's name "Bob" because
the AI's own first response had echoed the name - and that response survived the deletion window.

This PR fixes the example by inserting an unrelated conversation turn (a cat poem)
between the name introduction and the name recall query. This naturally pushes the
name-containing messages out of the window, so when the agent is asked "what's my name?"
it correctly responds that it doesn't know.

Type of change

[Update existing documentation]

Checklist

• I have read the contributing guidelines
• I have tested my changes locally using docs dev
• All code examples have been tested and work correctly

Related issues/PRs

closes #3894

Additional notes

The fix is minimal and surgical - only the example code and output block are changed.
No API behavior, no logic, no navigation changes. The updated output block is taken
from an actual local run to ensure accuracy.

[github-actions]

Thanks for opening a docs PR, Udit Sharma (@uditstocks)! When it's ready for review, please add the relevant reviewers:

• @npentrel or @lnhsingh (LangChain)