Fix MGUCell hidden state update

[lanarkite99]

What does this PR do?

Fixes #4608.

This fixes the nn.MGUCell hidden-state update equation to match the Minimal Gated Unit recurrence.

The current implementation computes:

h' = (1 - f) * n + f * h

but according to the MGU formulation, the gate should mix the previous hidden state and candidate activation as:

h' = (1 - f) * h + f * n

This PR updates the implementation and the displayed equation in the MGUCell docstring.

It also expands the existing MGU regression test so the cell output is checked against a manually computed recurrence for both:

reset_gate=True
reset_gate=False

Tested with:

python -m pytest tests/linen/linen_test.py -k mgu -q

Result:2 passed, 99 deselected

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.