Hubbard model notebook is flaky

[fdmalone]

From #869 we see some regression with FQE in the sense that the Taylor series might not converge. It's not clear when this broke or if it's a flaky numerical thing (too tight a threshold). There are a few options:

1. Update FQE to accept a threshold
2. Figure out if this is flaky or some external library update somewhere which lead to worsened precision.

At the very least we should test OUR notebooks where possible as part of some nightly tests.