Fix data initialization

[BenjaminRodenberg]

closes #80

This PR provides a draft (or a solution?) for #80.

Main changes

• Include initialize_data in initialize
• Make write_function optional argument of initialize
• Raise assertion, if no write_function is provided, but self._interface.is_action_required(precice.action_write_initial_data())
• We need some additional logic to be able to call coupling_expression.is_scalar_valued() or coupling_expression.is_vector_valued() on an empty coupling_expression. Empty means, created, but not updated.

Restrictions

• we do not check self._interface.is_read_data_available() anymore.

Todo

• delete prototype folder before merge
• update https://github.com/precice/tutorials accordingly. See Update FEniCS cases data initialization tutorials#101

[BenjaminRodenberg]

I highlighted some of the obvious todos of the current draft.

[IshaanDesai]

Changes align the adapter to a better workflow for initialization 👍
Some open points to be discussed and implemented (see comments above)

[BenjaminRodenberg]

Some comments from the discussion with @IshaanDesai and @uekerman.

We decided not add an self._interface.is_read_data_available() branch in read_data, since this will interfere with the CouplingExpression and PointSource update/creation.

In write_data we could optionally add the self._interface.is_write_data_required().

[IshaanDesai]

As we no longer return data from initialization and also initialize an empty coupling_expression we need slightly more QN-iterations in each time window than previous reference results. This result is known and was observed during the design overhaul.

New implementation:

bcs = [DirichletBC(V, u_D, boundary)]
coupling_expression = precice.create_coupling_expression()
bcs.append(.., coupling_expression)

while precice._is_coupling_ongoing:
   ...

Old implementation:

initial_data = read_data from preCICE before coupling starts
bcs = [DirichletBC(V, u_D, boundary)]
coupling_expression = precice.create_coupling_expression(initial_data)
bcs.append(.., coupling_expression)

while precice._is_coupling_ongoing:
   ...

[BenjaminRodenberg]

... we need slightly more QN-iterations in each time window than previous reference results.

The reason for this is probably (and hopefully) 3cd2fb8, where I intentionally removed initialization to see 1) whether our new scheme can handle cases without initialization and 2) what happens with the number of QN iterations. I would not expect a regression from this PR. I will now prepare a PR to update the tutorials according to this PR and carefully check for any regressions.

As soon as the tutorials PR is ready I would suggest to merge everything (don't forget to remove the heat.py from this PR again, since it was only introduced for prototyping).

[BenjaminRodenberg]

I prepared precice/tutorials#101 and checked that everything is working. I did not find anything that we have to worry about. So from my perspective this PR is ready to merge.