`document_ttl` squash logic may never be invoked for busy notebooks

[krassowski]

Description

document_ttl compares the time elapsed since the previous update was sent, thus often preventing squashing the updates for more active notebooks for a long time.

Reproduce

1. Install ipylab
2. Run jupyter lab --SQLiteYStore.document_ttl=1
3. Create a notebook with one cell:

from ipylab import JupyterFrontEnd
from time import sleep

app = JupyterFrontEnd()

for i in range(1000):
    app.commands.execute('notebook:insert-cell-below')
    sleep(0.1)
    app.commands.execute('notebook:delete-cell')
    sleep(0.1)

4. See that condition diff > self.document_ttl is never true, thus the squashing logic never gets executed:
   https://github.com/jupyter-server/pycrdt-websocket/blob/3a5648dc937baca71eb9e997a6aaf8e3d6c91654/pycrdt_websocket/ystore.py#L489

Expected behavior

We have a way to prevent SQL store from blowing to dangerous sizes, even if user continues to edit it. For example, if we set --SQLiteYStore.document_ttl=60 we expect to be able to have no more than 1 minute buffer in the history.

Context

pycrdt                    0.12.11
pycrdt-websocket          0.15.4
jupyterlab                4.4.0
jupyter-collaboration     4.0.0